Ultrasound-based radiogenomics: status, applications, and future direction

Article information

Ultrasonography. 2025;44(2):95-111

Publication date (electronic) : 2024 December 12

doi : https://doi.org/10.14366/usg.24152

Si-Rui Wang ¹^,^*

, Yu-Ting Shen ¹^,^*

, Bin Huang ²

, Hui-Xiong Xu^,¹

¹Department of Ultrasound, Institute of Ultrasound in Medicine and Engineering, Shanghai Institute of Medical Imaging, Zhongshan Hospital, Fudan University, Shanghai, China

²Department of Ultrasound, Zhejiang Hospital, Hangzhou, China

Correspondence to: Hui-Xiong Xu, MD, PhD, Department of Ultrasound, Institute of Ultrasound in Medicine and Engineering, Zhongshan Hospital, Fudan University, Shanghai 200032, P. R. China Tel, Fax. +86-21-66307539 E-mail: xuhuixiong@126.com

These authors contributed equally to this work.

Received 2024 August 12; Accepted 2024 December 12.

Abstract

Radiogenomics, an extension of radiomics, explores the relationship between imaging features and underlying gene expression patterns. This field is instrumental in providing reliable imaging surrogates, thus potentially representing an alternative to genetic testing. The rapidly growing area of radiogenomics that utilizes ultrasound (US) imaging seeks to elucidate the connections between US image characteristics and genomic data. In this review, the authors outline the radiogenomics workflow and summarize the applications of US-based radiogenomics. These include the prediction of gene variations, molecular subtypes, and other biological characteristics, as well as the exploration of the relationships between US phenotypes and cancer gene profiles. Although the field faces various challenges, US-based radiogenomics offers promising prospects and avenues for future research.

Keywords: Imaging genomics; Ultrasonography; Radiomics; Precision medicine; Neoplasms

Introduction

Medical imaging, a traditional diagnostic method, has become increasingly important in personalized precision medicine and clinical decision-making [1,2]. Recent advancements in image informatics technology have enabled the high-throughput extraction of quantitative data from digital medical images, prompting the emergence of the field of radiomics. The formal framework of radiomics, first proposed in 2012 [3], involves the extraction of numerous features from images using high-throughput computing and transforming them into mineable data [4]. Furthermore, progress in "omics" technologies, such as genomics, transcriptomics, and proteomics, has facilitated the comprehensive study of various aspects of biological systems [5]. Since the completion of the Human Genome Project, genomic information has been instrumental in uncovering the genetic variations that underlie human diseases, revolutionizing diagnosis, prognosis, and the prediction of treatment responses [6].

Broadly speaking, imaging genomics, also known as radiogenomics, investigates the relationships between imaging features and genetic markers, such as gene mutations, patterns of gene expression, and other relevant traits [7-9]. These imaging features may be qualitative or quantitative, with the latter often involving high-throughput radiomics. However, the definition of radiogenomics has narrowed in recent years due to advancements in radiomics research (Fig. 1) [3,7,9-14]. Radiogenomics, employing radiomics methodologies, is now specifically recognized as an application that studies correlations between quantitative imaging features and genomic characteristics [15,16]. Additionally, the term "radiogenomics" can refer to the investigation of genetic germline variations associated with normal tissue toxicity following radiotherapy [17]. In this review, "radiogenomics" is used in accordance with the first interpretation. In recent years, radiogenomics has demonstrated potential for rapid and non-invasive genotyping across various cancer types by uncovering links between the quantitative imaging characteristics of tumor phenotypes and genomic signatures. This has contributed to the shift toward precision medicine [12,18].

Fig. 1.

Historical evolution of the concept of radiogenomics.

The blue and orange boxes represent two components that are associated in radiogenomics studies. The blue boxes refer to imaging data and the orange boxes to genomics data. MRI, magnetic resonance imaging; CT, computed tomography; PET, positron emission tomography.

The advancement of artificial intelligence (AI) technologies, including machine learning (ML) and deep learning (DL), has demonstrated benefits in the detection, diagnosis, and classification of diseases [19-21]. AI has also been employed in radiogenomics for image processing and the development of powerful models that aid in making optimal clinical decisions [22].

The primary imaging modalities utilized in radiogenomics include magnetic resonance imaging (MRI), computed tomography (CT), positron emission tomography (PET), and ultrasound (US). US, a key area of medical imaging, offers distinct advantages such as the absence of ionizing radiation, portability, real-time imaging capability, and cost-effectiveness. Meanwhile, AI-based radiomics has considerably improved the accuracy and efficiency of US diagnostics by providing objective and quantitative assessments of US images [23,24]. US-based radiogenomics is an emerging field that relates US imaging features with gene expression patterns or genetic mutations. This area of research has been garnering increased research interest and shows considerable promise.

In this review, the authors present a brief overview of the US-based radiogenomics workflow and summarize pertinent research regarding thyroid, breast, liver, prostate, ovarian, and brain tumors. Furthermore, they examine the limitations and current challenges facing US-based radiogenomics and discuss its potential future applications.

Overview of Radiogenomics Workflows

US-based radiogenomics studies typically integrate the methodologies of US-based radiomics and genomics. Quantitative features are extracted, processed, and analyzed from US images and then associated with specific outcomes. Radiomics methods can be categorized as either classic/conventional radiomics, which involves extracting handcrafted features using conventional ML, or DL radiomics, which employs fully automated data analysis pipelines. The workflow of a radiogenomics study, specifically combining conventional radiomics and genomics, can generally be described in four steps (Fig. 2): (1) image and data acquisition, (2) image segmentation, (3) feature extraction and selection, and (4) data analysis and modeling for radiogenomics outcomes.

Fig. 2.

Applications of ultrasound (US)-based radiogenomics.

Genomic features are associated with ultrasound imaging features in tumors of the brain, thyroid, breast, liver, prostate, and ovary. IDH, isocitrate dehydrogenase; TERTp, telomerase-reverse transcriptase gene promoter; 1p/19q, chromosome arms 1p and 19q; ER, estrogen receptor; PR, progesterone receptor; HER2, human epidermal growth factor receptor 2.

(1) Image and data acquisition: The workflow begins with radiologists using consistent acquisition parameters to scan and obtain images of lesions. A range of modalities for US imaging can be utilized, including B-mode US (BUS), contrast-enhanced ultrasound (CEUS), color Doppler flow imaging (CDFI), and ultrasound elastography (UE). The selected high-quality US images are then recorded and stored in Digital Imaging and Communications in Medicine format. Adherence to standardized imaging protocols is crucial during data collection to minimize unwanted variability that could confound the results [25]. On the genomic front, molecular pathology data are acquired through genomic analyses such as microarray, DNA and RNA sequencing, or immunohistochemistry (IHC) techniques.

(2) Image segmentation: Segmentation is a critical step in the workflow because the extracted features depend upon the defined region of interest (ROI). Techniques for delineating the ROI include automatic, semi-automatic, and manual segmentation. In US-based radiogenomic studies, the ROI is most frequently demarcated manually (Table 1). Although manual segmentation by experienced radiologists is widely practiced, it is acknowledged to be time-consuming and subject to inter-observer variability. In contrast, automatic segmentation, which requires minimal operator intervention, provides more accurate and reproducible boundaries [33]. Semi-automatic segmentation may represent an effective compromise, allowing the expert to refine the output of the algorithm [34].

Table 1.

Overview of the US-based radiogenomics literature on thyroid cancer

(3) Feature extraction and selection: An essential step in the radiogenomics process is the extraction of high-dimensional features from defined ROIs to characterize phenotypes. The extraction of US radiomics features from ROIs is typically performed using specialized software, such as PyRadiomics (https://pyradiomics.readthedocs.io/). Extracted features are primarily categorized into shape-based features, intensity-based (first-order) features, texture (second-order) features, higher-order features, and additional features based on filtering and transformation [4]. The Image Biomarker Standardization Initiative (IBSI) provides guidelines that are recommended for detailed descriptions of radiomics features and their corresponding equations [35]. In the process of feature extraction, hundreds or thousands of features may be computed. Due to potential redundancy among these features, it is crucial to select those that capture essential information to avoid overfitting. Techniques such as least absolute shrinkage and selection operator (Lasso) regression and maximum relevance minimum redundancy are often used to aid in choosing the most informative features by reducing the dimensionality of the dataset.

(4) Modeling for outcomes: Subsequently, ML techniques can be utilized to relate the extracted image features with genomic data and to build predictive models. Conventional ML methods, including logistic regression, random forest, support vector machine (SVM), and eXtreme Gradient Boosting (XGBoost) are often applied in the development of models that predict cancer molecular subtypes, gene mutations, or gene expression patterns. The performance of these models, derived from the training set, should be validated or tested on a separate testing set using evaluation metrics such as receiver operating characteristic curve, calibration curve, and clinical decision curve analysis. External validation is considered more credible than internal validation, as the results are typically more robust when validated in independent datasets [2].

DL models are based on a number of advanced neural networks, including deep convolutional neural networks (CNNs), transformers, and other deep neural networks. DL can be integrated into radiomics or genomics research, enabling models to automatically process medical imaging or genomic data as input and produce outcomes that align with the objectives of the study. In radiogenomics studies, DL can serve as a method for association in the mapping process or can be applied to create new model architectures that utilize both imaging data and molecular data.

Application

US-based radiogenomics is still in its nascent phase, with investigations focused on a limited number of cancer types, including those of the thyroid, breast, liver, prostate, ovary, and brain. This section examines US-based radiogenomics research that seeks to uncover correlations between radiomics signatures and gene expression profiles (GEPs) or biological pathways. Additionally, it discusses studies aiming to predict genetic alterations and molecular subtypes across various cancers (Fig. 3).

Fig. 3.

General workflow of ultrasound (US)-based radiogenomics studies.

Initially, both US image and gene-related data are collected from patients and processed. Subsequently, machine learning or deep learning algorithms are employed to integrate radiomics data with genomic or other “omics” data to construct a model. The results of the radiogenomics analysis are then utilized to aid in patient diagnosis or prognosis.

Thyroid

Genomic alterations play a pivotal role in the development of thyroid cancer, with driver mutations in genes such as BRAF, RAS, and RET influencing the clinicopathological characteristics of thyroid tumors [36]. Consequently, molecular characterization of thyroid cancer is instrumental in more effectively identifying various subtypes of thyroid carcinoma, each exhibiting distinct clinical behaviors and responses to radioactive iodine and targeted therapies [37,38]. US-based radiogenomics could potentially assist in the diagnosis, prognosis, and treatment planning of thyroid cancer through non-invasive methods.

Prediction of BRAF V600E Gene Mutation

A strong association has been demonstrated between the BRAF V600E mutation and the development or progression of papillary thyroid carcinoma (PTC), implicating BRAF V600E as a critical driver of the PTC phenotype [39].

Two preliminary studies attempted to assess the potential of BUS radiomics for predicting BRAF mutation status in PTC, but the outcomes were not encouraging [26,27]. Yoon et al. [26] collected preoperative US images to extract features and calculate a radiomics score, concluding that US-based radiomics did not accurately predict the BRAF V600E mutation in PTCs. Similarly, Kwon et al. [27] reported only moderate performance of classifier models using US radiomics to predict the presence of BRAF mutations in PTCs.

In contrast, two studies from a single team have demonstrated the usefulness of UE images in predicting genetic mutations [28,29]. Wang et al. [28] found that combining UE radiomics features with grayscale US radiomics features demonstrated clinical utility in predicting BRAF V600E mutations among patients with PTC. Building on these findings, Agyekum et al. [29] employed UE radiomics features to develop six distinct ML models. They evaluated the performance of these models in predicting the BRAF V600E mutation and reported that the UE-based radiomics models performed well, with an area under the curve (AUC) ranging from 0.80 to 0.98. These studies suggest that multimodal US could address the limitations of conventional US in extracting gene expression information from target organs.

Prediction of RET Rearrangement

The RET ("rearranged during transfection") gene, a known oncogenic driver in thyroid cancer, is responsible for chromosomal rearrangements such as the RET/PTC fusion gene. Recent clinical evidence suggests that RET rearrangements occur in 10% to 20% of PTCs [40].

Yu et al. [30] developed a DL radiomics nomogram utilizing US to preoperatively predict RET rearrangement in patients with PTC. The nomogram model that integrated clinical and DL radiomics signatures outperformed models based solely on clinical or radiomics signatures, achieving an AUC of 0.9545 in the test cohort and an AUC of 0.9396 in the training cohort. The study utilized a DL methodology to extract US image features imperceptible to the human eye, underscoring the utility of DL in the construction of predictive models for radiogenomics.

Analysis of Gene Expression Levels

Tong et al. [31] conducted a radiogenomics analysis in patients with PTC to predict cervical lymph node metastasis and to explore the relationship between US radiomics features and the molecular characteristics of PTC. Radiomics features and gene modules, obtained through a clustering approach, were correlated to construct a radiogenomic map. This map revealed nine statistically significant pairwise correlations, indicating potential radiogenomic biomarkers for PTC (Fig. 4).

Fig. 4.

Radiogenomics analysis of radiomics features associated with gene modules.

A radiogenomic correlation map is generated using the Spearman rank correlation method to assess the relationships between ultrasound radiomics features and gene modules. Nine pairwise correlations marked with an asterisk denote statistically significant associations (P<0.05). Red indicates positive correlations, while blue denotes negative correlations. *Statistically significant association. Reprinted from Tong et al. Front Oncol 2021;11:682998 [31], according to Creative Commons License.

Lee et al. [32] analyzed US features assessed by CNNs and radiologists, along with the associated molecular biological mechanisms. They found that certain suspicious US features, such as marked hypoechogenicity, non-circumscribed margins, microcalcifications, and nonparallel shape, were indicative of the upregulation of genes linked to immune response and epithelial-mesenchymal transition. The research suggested that the US characteristics of PTCs, as evaluated by both radiologists and CNNs, could potentially predict the biological behavior and molecular features of the tumors. These radiogenomics studies establish a correlation between radiomics features and complex gene expression data while exploring the underlying molecular mechanisms, this supporting the role of precision medicine in cancer management.

Among US-based radiogenomics studies concerning thyroid cancer, most have investigated the potential of radiomics features to predict specific gene variations in PTC. Additionally, some studies have focused on the correlation between radiomics features and gene expression patterns. The principal findings of these radiogenomics studies are summarized in Table 1.

Breast

Breast cancer represents a significant health risk, as it is the leading cause of cancer-related mortality among women worldwide [41]. Since the first article on breast cancer radiogenomics was published in 2012 [42], a growing body of research has been produced on the subject, utilizing various imaging modalities [43]. The advanced study of genomic markers in breast cancer enables analyses that range from identifying surrogate molecular subtypes to conducting multigene expression-based assays [43]. In the context of breast cancer management, US-based radiogenomics could potentially be employed to predict molecular subtypes and examine the relationships between US characteristics and gene expression, thus providing a clearer picture of the behavior and progression of breast cancer (Tables 2, 3).

Table 2.

Overview of the US-based radiogenomics literature on breast cancer (molecular subtypes)

Table 3.

Overview of the US-based radiogenomics literature on breast cancer (BRCA mutation and gene expression)

Prediction of Molecular Subtypes

Based on gene expression profiling, breast cancer can be classified into four molecular subtypes: luminal A, luminal B, human epidermal growth factor receptor 2 (HER2)–enriched, and basal-like [52,53]. However, identifying molecular subtypes at the genetic level can be expensive and time-consuming. Consequently, in both clinical practice and research, the use of IHC surrogates for breast cancer subtyping is becoming increasingly common [43]. Breast cancers can be categorized into St. Gallen molecular subtypes according to IHC detection of estrogen receptor (ER), progesterone receptor (PR), HER2, and the Ki-67 labeling index, which correlate with intrinsic subtypes [54,55]. Different molecular subtypes demonstrate variations in disease presentation, prognosis, and response to treatment. The primary goal of US-based radiogenomics research in breast cancer is to predict molecular subtypes by identifying specific US imaging features.

In an earlier study, Guo et al. [44] employed a US-based radiomics approach to differentiate between hormone receptor-positive, HER2- negative cancers and triple-negative breast cancers (TNBC). They demonstrated that quantitative US characteristics are strongly linked to receptor status and molecular subtypes in breast cancer. Xu et al. [45] developed four SVM models capable of effectively predicting the molecular expression of ER, PR, HER2, and Ki-67 in breast cancer. To distinguish TNBC from non-TNBC prior to surgery, Xu et al. [46] created a US-based radiomics nomogram that combines US radiomics features with conventional US features, achieving an AUC value of 0.813 in the validation set.

US-based DL algorithms have shown promising results in the prediction of molecular subtypes of breast cancer. Jiang et al. [47] conducted a multicenter retrospective study demonstrating that a CNN model, developed using US images, was highly accurate in distinguishing between luminal and non-luminal diseases, as well as in diagnosing molecular subtypes. Zhou et al. [48] evaluated the performance of three assembled CNN (ACNN) models in predicting molecular subtypes of breast cancer. Their findings indicated that the multimodal ACNN, which integrated grayscale US images, CDFI, and shear wave elastography (SWE), was effective in accurately predicting these subtypes, with an AUC ranging from 0.87 to 0.96. Quan et al. [49] found that a DL radiomics approach, which incorporated breast US video and clinical characteristics, showed superior performance in predicting HER2 expression status, achieving an AUC of 0.81.

Gong et al. [50] conducted pioneering research on the utility of CEUS radiomics for predicting molecular subtypes of breast cancer. They integrated CEUS video data with conventional US images to develop a radiomics model capable of classifying breast cancer into six molecular subtypes. Their findings revealed that CEUS radiomics offers a distinct advantage in determining the receptor status of breast cancer, as the CEUS videos capture dynamic blood perfusion details that reflect vascular features of the tumor. This work also underscores the important role of CEUS in the field of radiogenomics.

Based on imaging features from US and mammography, as well as clinical characteristics, Ma et al. [62] developed an interpretable ML model that demonstrated good performance in identifying breast cancer. Furthermore, it assisted radiologists in distinguishing TNBC from other subtypes, exhibiting an AUC of 0.971. This predictive model incorporated two primary imaging techniques—US and mammography—in breast cancer diagnosis and yielded excellent results. It provides valuable insights into the potential benefits of integrating additional imaging modalities in radiogenomics studies.

Prediction of BRCA Mutation

Germline pathogenic variants in the breast cancer susceptibility genes BRCA1/2 are associated with an increased risk of developing ovarian or breast cancer [63]. Guo et al. [56] developed a nomogram model that integrates radiomics derived from US with clinical characteristics, achieving a high accuracy in detecting germline BRCA1/2 (gBRCA1/2) mutations, as indicated by an AUC of 0.811. Similarly, Deng et al. [51] employed radiomics data from both intratumoral and peritumoral regions, in conjunction with clinicopathological factors, to construct a predictive nomogram. This model demonstrated an AUC of 0.824 for the accurate identification of gBRCA mutations. When a patient has a high risk of carrying a BRCA pathogenic variant, genetic testing is typically recommended [64]. US-based radiogenomics may offer a cost-effective approach for the diagnosis of breast cancer and the identification of individual BRCA mutations, which can inform personalized treatment strategies.

Analysis of Gene Expression

Cui et al. [57] developed a predictive model by conducting a thorough analysis of US radiomics features of HER2-positive breast cancer, exploring the associated gene expression and biological information. Utilizing mRNA expression profiles for HER2-positive breast cancer from the Gene Expression Omnibus (GEO) database, they found that various US radiomics features corresponded to different biological functions. Specifically, one of the US radiomics features, zone entropy, which is elevated in US images of HER2-positive breast cancer, may signify immune cell activity (Fig. 5). This study demonstrates that US radiomics features can reflect the biological functions of cancer in a differential manner.

Fig. 5.

A feature-mRNA-function network constructed using an independent dataset.

The hierarchical model reveals relationships among five differential ultrasound radiomics features, specific genes, and particular functions in human epidermal growth factor receptor 2 (HER2)–positive breast cancer. The analysis indicates that the size zone non-uniformity feature correlates with carbon metabolism in cancer. The size zone non-uniformity normalized feature is associated with cell-substrate junctions. Small area emphasis has a connection to chemical carcinogenesis, while the small area high gray level emphasis feature relates to the composition of the endomembrane system. Lastly, zone entropy is linked to the activities of peroxidase and oxidoreductase. Reprinted from Cui et al. J Transl Med 2023;21:44 [57], according to Creative Commons License.

In a study by Huang et al. [58], a model was developed that integrated molecular subtypes with US features to predict TP53 and PIK3CA gene mutations in breast cancer. The findings indicated that larger tumor size and a mass-like appearance on US were associated with PIK3CA mutation, while calcifications on US were an independent predictor of TP53 mutation. In a later study [59], the researchers evaluated the correlation between US radiomics features and genomic sequencing in patients with and without TNBC. The sequencing of breast cancer-specific genes revealed that US radiomics features were linked to gene mutations and signaling pathways that differed between the TNBC and non-TNBC groups. These findings suggest that a radiogenomic signature based on US could potentially serve as a predictor for molecular targets in the precision treatment of TNBC.

Using BUS and vascular US (including microvascular US and CEUS) imaging of breast cancers, Park et al. [60] investigated the relationship between US morphological and vascular characteristics and genetic alterations, as revealed by RNA sequencing results from 31 patients with breast cancer. In this radiogenomic study, 26 genes associated with tumor growth, prognosis, and metastasis in breast cancer were found to be significantly upregulated or downregulated in relation to US phenotypes. A heatmap was created to display the gene expression data in breast cancers with nonparallel and parallel orientations (Fig. 6). Building on these findings, the researchers also analyzed DNA sequencing results from the same 31 breast cancer samples. They found that certain vascular features observed on US were significantly associated with genetic alterations linked to angiogenesis and the prognosis of breast cancer [61]. These studies underscore the potential of vascular US to provide surrogate markers for treatment decisions and prognostic predictions in patients with breast cancer.

Fig. 6.

Heat map illustrating 42 differentially expressed genes in relation to the orientation of breast cancer on B-mode ultrasound (BUS) imaging.

The columns in the table represent 31 individual breast cancer cases, with yellow indicating parallel orientation and green representing nonparallel orientation. The rows correspond to the 42 differentially expressed genes. The color key reflects the level of gene expression: red signifies upregulation and blue signifies downregulation. In the BUS image, a 53-year-old patient with parallel-oriented breast cancer is highlighted with a yellow border, and a 48-year-old patient with nonparallel-oriented breast cancer is highlighted with a green border. Reprinted from Park et al. Radiology 2020;295:24-34 [60], with permission from the Radiological Society of North America Inc.

Liver

Primary liver cancers are a major contributor to cancer-related mortality worldwide [41]. Although limited US radiogenomics and radiomics studies have focused on the liver, the applications of US radiomics in liver tumors have been delineated in terms of diagnosis [65,66] and the prediction of microvascular invasion [67], as well as in assessing the biological behaviors of liver tumors [68,69]. Given the diverse array of treatment options for liver cancers, radiogenomics holds promise for improving the clinical management of these malignancies, including hepatocellular carcinoma (HCC).

Wang et al. [70] developed a radiomics model that leverages CEUS features from the arterial, portal venous, and delayed phases to predict the T cell–inflamed GEP of 18 genes in patients with HCC (Table 4). The ML model, which utilized features from the US arterial phase, exhibited excellent performance in predicting T cell–inflamed GEP, achieving an average AUC of 0.905 in 5-fold cross-validation. This pioneering study in radiogenomics demonstrated the potential of CEUS radiomics to predict T cell–inflamed GEP, indicative of a T cell–activated tumor microenvironment, thus offering valuable insights that could inform therapeutic decisions in immunotherapy.

Table 4.

Overview of the US-based radiogenomics literature in liver, prostate, ovarian, and brain tumors

Prostate

Prostate cancer represents the second most common cancer among men, exhibiting higher incidence and mortality rates in the Western world [71]. Several US-based radiomics studies have analyzed quantitative data from US images or radiofrequency (RF) signals, demonstrating their potential utility in detecting [72,73] and localizing [74] prostate cancer, as well as predicting biochemical recurrence [75].

By characterizing cancer-associated fibroblasts (CAFs), Ageeli et al. [76] explored the prostate tumor microenvironment. They correlated tissue stiffness measurements obtained from transrectal US SWE with IHC and genomics techniques. Their findings demonstrated a positive correlation between US SWE–measured tissue stiffness and the expression of smooth muscle actin α and platelet-derived growth factor receptor β in CAFs (Table 4). In another recent study, proteomic data were linked to qualitative features of BUS and CEUS in prostate cancer, revealing the biological functions that underpin the ultrasonic phenotype [77]. These radiogenomic insights suggest that each US imaging modality, whether SWE or CEUS, could serve as a non-invasive tool for visualizing the tumor microenvironment. This capability improves our understanding of the biological mechanisms at play in prostate cancer and may inform personalized treatment strategies.

Ovary

Pelvic US scans are routinely utilized in clinical settings for gynecological examinations. Women with gBRCA1 mutations are at a heightened risk of developing ovarian cancer, one of the deadliest cancers affecting women [80].

To predict gBRCA1/2 status in women with healthy ovaries, Nero et al. [78] developed an automated ML model using features extracted from transvaginal pelvic US images based on real-world data (Table 4). This model could help identify patients more likely to carry gBRCA1/2 pathogenic variants, thus reducing the need for unnecessary genetic testing. By integrating this model with existing clinical and family history criteria, they also conducted a cost-effectiveness analysis comparing various methods of identifying BRCA1/2 carriers in the general population [81]. Their findings suggest that the incidence of BRCA-related cancers could be significantly lowered by offering genetic testing to women selected through the radiogenomic model. The model offered an effective, cost-efficient, and sustainable approach to preventing BRCA-related cancers, setting the stage for its future incorporation into cancer management protocols.

Brain

Gliomas constitute the majority of malignant brain tumors [82]. In line with the World Health Organization Classification of Tumors of the Central Nervous System guidelines, molecular parameters are now incorporated as biomarkers for the accurate classification of gliomas [83]. Radiogenomics is an intriguing field that hold great promise for improving the management of gliomas via personalized medicine [84].

US RF signals are echo signals produced by the interaction of US with human tissues. Xie et al. [79] developed a DL modeling method that leveraged the temporal-spatial characteristics of US RF signals, leading to an effective intraoperative diagnosis of glioma biomarkers, including isocitrate dehydrogenase 1, chromosome arms 1p and 19q, and telomerase-reverse transcriptase gene promoter (Table 4). The spatial-temporal integration model derived from US RF signals outperformed two classical models in diagnosing molecular biomarkers, achieving a mean AUC of 0.96. This model thus provides accurate diagnoses and predictions of prognosis. While many studies have focused on achieving preoperative molecular diagnosis using medical imaging modalities such as CT and MRI [85], their use during surgery is limited. The US-based method offers a rapid intraoperative molecular detection solution that better meets clinical needs for personalized medicine.

Challenges and Future Direction

Radiogenomics plays a pivotal role in the realm of precision medicine for cancer, offering clinicians valuable insights to inform cancer detection and treatment strategies. This development is a natural progression in the shift toward precision medicine, with radiogenomics not only providing a cost-effective alternative to traditional gene sequencing but also yielding information about the entire tumor, as opposed to a limited biopsy specimen. Radiogenomics studies primarily utilize MRI, CT, and PET to study a variety of tumors, including those in the brain, lung, breast, liver, prostate, kidney, colorectum, and gynecological areas [86]. US-based radiogenomics is still in its nascent phase, with research largely focusing on the relationships between imaging features and specific genes or molecular subtypes, particularly in thyroid and breast cancers. Nonetheless, US-based radiogenomics offers several distinct benefits: it is radiation-free, time-efficient, and cost-effective. The capacity to perform real-time scanning and rapid diagnosis has allowed US to be integrated into the surgical workflow, thus aiding surgeons. Moreover, image resolution and quality have been improved by advancements in the latest US technologies. Various US imaging modalities, such as CEUS and SWE, deliver multifaceted information regarding lesions, thus optimizing disease diagnosis. In this section, the authors explore the challenges encountered by US-based radiogenomics and highlight potential future directions.

Compared to automated image acquisition in radiogenomics using MRI or CT, US imaging demonstrates poor objectivity and reproducibility. This is largely due to the high operator dependence in the acquisition and evaluation of US images, which presents challenges in maintaining consistent image quality within US-based radiogenomics workflows. Furthermore, the complex nature of the radiomics workflow—which includes collecting images, delineating ROIs, extracting optimal features, and building appropriate models—contributes to variability in results. The absence of standardized processes has resulted in a lack of repeatability and reproducibility in radiogenomic studies [87]. To address these challenges, multiple international consortiums have been established. The proposed radiomics quality score aims to increase the robustness of radiomics research, while the IBSI was developed to provide a standardized approach for feature computation and image processing in radiomics analysis [1,35]. Consequently, efforts should be made to establish uniform standards for US operation and image processing procedures to ensure the production of high-quality and reproducible studies.

In contrast to the substantial and publicly accessible datasets found in other medical imaging fields, the availability of public datasets in medical US is currently limited. This scarcity of publicly available databases hampers researchers’ ability to conduct studies with large sample sizes [88]. The Cancer Genome Atlas (TCGA) data, which includes genomic and clinical biomarkers from various cancer types, can be linked to corresponding imaging data available for cancer research in The Cancer Imaging Archive (TCIA). The use of TCGA-TCIA data has been instrumental in advancing radiogenomics research that utilizes MRI, CT, and PET imaging modalities [89,90]. The development of large-scale imaging, genomics, or combined datasets as a resource for research teams worldwide could greatly improve the quality of US-based radiogenomics studies and accelerate their integration into clinical practice.

The integration of various US imaging modalities in multimodal US fusion analysis enables thorough evaluation of the imaged tissue. Multimodal radiomics has demonstrated substantial progress in improving clinical diagnostic accuracy [15]. Another key imaging modality in medical imaging is three-dimensional (3D) US, which offers considerable potential in US-based clinical applications and provides more information than two-dimensional images. Cutting-edge 3D AI algorithms are anticipated to play a pivotal role in radiogenomics based on US, as these are designed to perform a variety of tasks in medical US analysis [91].

While US-based radiogenomics is currently limited to a few cancer types, the scope of radiomics utilizing US has expanded to include investigations into gallbladder cancer [92], renal cancer [93], rectal cancer [94], endometrial cancer [95], soft-tissue tumors [96], and others. Thus, it is reasonable to anticipate broader applications of US-based radiogenomics in as-yet-unexplored areas. Moreover, radiogenomics research that employs MRI and CT imaging has demonstrated potential as an effective tool for predicting therapeutic responses and assessing recurrence risk. This suggests a promising avenue for further research into US-based radiogenomics, particularly in its capacity to correlate imaging characteristics with patient outcomes and genomic data.

When designing a radiogenomics study, researchers must recognize the importance of standardized protocols, as decisions made at each stage will impact subsequent steps in the radiogenomics workflow. Furthermore, it is advisable for future radiogenomics research to prioritize prospective, multicenter studies with large sample sizes to improve the robustness of the research and drive progress in this field.

Conclusion

In this review, we discuss the current state of radiogenomics studies that utilize US and explore the challenges and prospects of this field. US-based radiogenomics research is still in its infancy, with only a limited number of studies conducted thus far. However, it represents an emerging and promising area for genotype identification due to its wide availability, cost-effectiveness, and non-invasive nature. To address the identified challenges, future US-based radiogenomics research will require high-quality, multicenter, and multi-geographical studies with a standardized workflow. With the expansion of clinical data and advancements in AI approaches, further investigation in this rapidly expanding field will be crucial for the successful integration of radiogenomics into clinical practice.

Notes

Author Contributions

Conceptualization: Wang SR, Huang B, Xu HX. Data acquisition: Wang SR, Shen YT. Data analysis or interpretation: Shen YT, Xu HX. Drafting of the manuscript: Wang SR, Shen YT. Critical revision of the manuscript: Huang B, Xu HX. Approval of the final version of the manuscript: all authors.

Conflict of Interest

No potential conflict of interest was reported by the authors.

Acknowledgments

This work was supported by the Ministry of Science and Technology (2023YFC2414204), the Shanghai Municipal Health Commission (Grant SHSLCZDZK 03502), the Science and Technology Commission of Shanghai Municipality (Grant 19DZ2251100), Fudan University (IDF152076), and Zhongshan Hospital of Fudan University (Grant 2022ZSQD07).

References

1. Lambin P, Leijenaar RT, Deist TM, Peerlings J, de Jong EE, van Timmeren J, et al. Radiomics: the bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol 2017;14:749–762.

2. Guiot J, Vaidyanathan A, Deprez L, Zerka F, Danthine D, Frix AN, et al. A review in radiomics: making personalized medicine a reality via routine imaging. Med Res Rev 2022;42:426–440.

3. Lambin P, Rios-Velazquez E, Leijenaar R, Carvalho S, van Stiphout RG, Granton P, et al. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer 2012;48:441–446.

4. Gillies RJ, Kinahan PE, Hricak H. Radiomics: images are more than pictures, they are data. Radiology 2016;278:563–577.

5. Olivier M, Asmis R, Hawkins GA, Howard TD, Cox LA. The need for multi-omics biomarker signatures in precision medicine. Int J Mol Sci 2019;20:4781.

6. Shendure J, Findlay GM, Snyder MW. Genomic medicine: progress, pitfalls, and promise. Cell 2019;177:45–57.

7. Mazurowski MA. Radiogenomics: what it is and why it is important. J Am Coll Radiol 2015;12:862–866.

8. Incoronato M, Aiello M, Infante T, Cavaliere C, Grimaldi AM, Mirabelli P, et al. Radiogenomic analysis of oncological data: a technical survey. Int J Mol Sci 2017;18:805.

9. Kuo MD, Jamshidi N. Behind the numbers: decoding molecular phenotypes with radiogenomics: guiding principles and technical considerations. Radiology 2014;270:320–325.

10. Rutman AM, Kuo MD. Radiogenomics: creating a link between molecular diagnostics and diagnostic imaging. Eur J Radiol 2009;70:232–241.

11. Sala E, Mema E, Himoto Y, Veeraraghavan H, Brenton JD, Snyder A, et al. Unravelling tumour heterogeneity using next-generation imaging: radiomics, radiogenomics, and habitat imaging. Clin Radiol 2017;72:3–10.

12. Jansen RW, van Amstel P, Martens RM, Kooi IE, Wesseling P, de Langen AJ, et al. Non-invasive tumor genotyping using radiogenomic biomarkers, a systematic review and oncology-wide pathway analysis. Oncotarget 2018;9:20134–20155.

13. Bodalal Z, Trebeschi S, Nguyen-Kim TD, Schats W, Beets-Tan R. Radiogenomics: bridging imaging and genomics. Abdom Radiol (NY) 2019;44:1960–1984.

14. Singh S, Mohajer B, Wells SA, Garg T, Hanneman K, Takahashi T, et al. Imaging genomics and multiomics: a guide for beginners starting radiomics-based research. Acad Radiol 2024;31:2281–2291.

15. Liu Q, Hu P. Extendable and explainable deep learning for pan-cancer radiogenomics research. Curr Opin Chem Biol 2022;66:102111.

16. Vaidya T, Agrawal A, Mahajan S, Thakur MH, Mahajan A. The continuing evolution of molecular functional imaging in clinical oncology: the road to precision medicine and radiogenomics (part I). Mol Diagn Ther 2019;23:1–26.

17. Andreassen CN, Schack LM, Laursen LV, Alsner J. Radiogenomics: current status, challenges and future directions. Cancer Lett 2016;382:127–136.

18. Liu Z, Duan T, Zhang Y, Weng S, Xu H, Ren Y, et al. Radiogenomics: a key component of precision cancer medicine. Br J Cancer 2023;129:741–753.

19. Trivizakis E, Papadakis GZ, Souglakos I, Papanikolaou N, Koumakis L, Spandidos DA, et al. Artificial intelligence radiogenomics for advancing precision and effectiveness in oncologic care (Review). Int J Oncol 2020;57:43–53.

20. Wang X, Jing L, Yan L, Wang P, Zhao C, Xu H, et al. A conditional inference tree model for predicting cancer risk of non-mass lesions detected on breast ultrasound. Eur Radiol 2024;34:4776–4788.

21. Zhu AQ, Wang Q, Shi YL, Ren WW, Cao X, Ren TT, et al. A deep learning fusion network trained with clinical and high-frequency ultrasound images in the multi-classification of skin diseases in comparison with dermatologists: a prospective and multicenter study. EClinicalMedicine 2024;67:102391.

22. Saxena S, Jena B, Gupta N, Das S, Sarmah D, Bhattacharya P, et al. Role of artificial intelligence in radiogenomics for cancers in the era of precision medicine. Cancers (Basel) 2022;14:2860.

23. Shen YT, Chen L, Yue WW, Xu HX. Artificial intelligence in ultrasound. Eur J Radiol 2021;139:109717.

24. Zhang H, Meng Z, Ru J, Meng Y, Wang K. Application and prospects of AI-based radiomics in ultrasound diagnosis. Vis Comput Ind Biomed Art 2023;6:20.

25. Sollini M, Antunovic L, Chiti A, Kirienko M. Towards clinical application of image mining: a systematic review on artificial intelligence and radiomics. Eur J Nucl Med Mol Imaging 2019;46:2656–2672.

26. Yoon JH, Han K, Lee E, Lee J, Kim EK, Moon HJ, et al. Radiomics in predicting mutation status for thyroid cancer: a preliminary study using radiomics features for predicting BRAFV600E mutations in papillary thyroid carcinoma. PLoS One 2020;15e0228968.

27. Kwon MR, Shin JH, Park H, Cho H, Hahn SY, Park KW. Radiomics study of thyroid ultrasound for predicting BRAF mutation in papillary thyroid carcinoma: preliminary results. AJNR Am J Neuroradiol 2020;41:700–705.

28. Wang YG, Xu FJ, Agyekum EA, Xiang H, Wang YD, Zhang J, et al. Radiomic model for determining the value of elasticity and grayscale ultrasound diagnoses for predicting BRAF(V600E) mutations in papillary thyroid carcinoma. Front Endocrinol (Lausanne) 2022;13:872153.

29. Agyekum EA, Wang YG, Xu FJ, Akortia D, Ren YZ, Chambers KH, et al. Predicting BRAFV600E mutations in papillary thyroid carcinoma using six machine learning algorithms based on ultrasound elastography. Sci Rep 2023;13:12604.

30. Yu J, Zhang Y, Zheng J, Jia M, Lu X. Ultrasound images-based deep learning radiomics nomogram for preoperative prediction of RET rearrangement in papillary thyroid carcinoma. Front Endocrinol (Lausanne) 2022;13:1062571.

31. Tong Y, Sun P, Yong J, Zhang H, Huang Y, Guo Y, et al. Radiogenomic analysis of papillary thyroid carcinoma for prediction of cervical lymph node metastasis: a preliminary study. Front Oncol 2021;11:682998.

32. Lee J, Yoon JH, Lee E, Lee HY, Jeong S, Park S, et al. Immune response and mesenchymal transition of papillary thyroid carcinoma reflected in ultrasonography features assessed by radiologists and deep learning. J Adv Res 2024;62:219–228.

33. Parmar C, Rios Velazquez E, Leijenaar R, Jermoumi M, Carvalho S, Mak RH, et al. Robust radiomics feature quantification using semiautomatic volumetric segmentation. PLoS One 2014;9e102107.

34. Pacchiano F, Tortora M, Doneda C, Izzo G, Arrigoni F, Ugga L, et al. Radiomics and artificial intelligence applications in pediatric brain tumors. World J Pediatr 2024;20:747–763.

35. Zwanenburg A, Vallieres M, Abdalah MA, Aerts H, Andrearczyk V, Apte A, et al. The image biomarker standardization initiative: standardized quantitative radiomics for high-throughput image-based phenotyping. Radiology 2020;295:328–338.

36. Landa I, Cabanillas ME. Genomic alterations in thyroid cancer: biological and clinical insights. Nat Rev Endocrinol 2024;20:93–110.

37. Boucai L, Zafereo M, Cabanillas ME. Thyroid cancer: a review. JAMA 2024;331:425–435.

38. Xu H, Zhang Y, Wu H, Zhou N, Li X, Pineda JP, et al. High diagnostic accuracy of epigenetic imprinting biomarkers in thyroid nodules. J Clin Oncol 2023;41:1296–1306.

39. Scheffel RS, Dora JM, Maia AL. BRAF mutations in thyroid cancer. Curr Opin Oncol 2022;34:9–18.

40. Subbiah V, Yang D, Velcheti V, Drilon A, Meric-Bernstam F. State-of-the-art strategies for targeting RET-dependent cancers. J Clin Oncol 2020;38:1209–1221.

41. Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A, et al. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2021;71:209–249.

42. Yamamoto S, Maki DD, Korn RL, Kuo MD. Radiogenomic analysis of breast cancer using MRI: a preliminary study to define the landscape. AJR Am J Roentgenol 2012;199:654–663.

43. Grimm LJ, Mazurowski MA. Breast cancer radiogenomics: current status and future directions. Acad Radiol 2020;27:39–46.

44. Guo Y, Hu Y, Qiao M, Wang Y, Yu J, Li J, et al. Radiomics analysis on ultrasound for prediction of biologic behavior in breast invasive ductal carcinoma. Clin Breast Cancer 2018;18:e335–e344.

45. Xu R, You T, Liu C, Lin Q, Guo Q, Zhong G, et al. Ultrasound-based radiomics model for predicting molecular biomarkers in breast cancer. Front Oncol 2023;13:1216446.

46. Xu M, Zeng S, Li F, Liu G. Utilizing grayscale ultrasound-based radiomics nomogram for preoperative identification of triple negative breast cancer. Radiol Med 2024;129:29–37.

47. Jiang M, Zhang D, Tang SC, Luo XM, Chuan ZR, Lv WZ, et al. Deep learning with convolutional neural network in the assessment of breast cancer molecular subtypes based on US images: a multicenter retrospective study. Eur Radiol 2021;31:3673–3682.

48. Zhou BY, Wang LF, Yin HH, Wu TF, Ren TT, Peng C, et al. Decoding the molecular subtypes of breast cancer seen on multimodal ultrasound images using an assembled convolutional neural network model: a prospective and multicentre study. EBioMedicine 2021;74:103684.

49. Quan MY, Huang YX, Wang CY, Zhang Q, Chang C, Zhou SC. Deep learning radiomics model based on breast ultrasound video to predict HER2 expression status. Front Endocrinol (Lausanne) 2023;14:1144812.

50. Gong X, Li Q, Gu L, Chen C, Liu X, Zhang X, et al. Conventional ultrasound and contrast-enhanced ultrasound radiomics in breast cancer and molecular subtype diagnosis. Front Oncol 2023;13:1158736.

51. Deng T, Liang J, Yan C, Ni M, Xiang H, Li C, et al. Development and validation of ultrasound-based radiomics model to predict germline BRCA mutations in patients with breast cancer. Cancer Imaging 2024;24:31.

52. Perou CM, Sorlie T, Eisen MB, van de Rijn M, Jeffrey SS, Rees CA, et al. Molecular portraits of human breast tumours. Nature 2000;406:747–752.

53. Sorlie T, Perou CM, Tibshirani R, Aas T, Geisler S, Johnsen H, et al. Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications. Proc Natl Acad Sci U S A 2001;98:10869–10874.

54. Goldhirsch A, Winer EP, Coates AS, Gelber RD, Piccart-Gebhart M, Thurlimann B, et al. Personalizing the treatment of women with early breast cancer: highlights of the St Gallen International Expert Consensus on the Primary Therapy of Early Breast Cancer 2013. Ann Oncol 2013;24:2206–2223.

55. Goldhirsch A, Wood WC, Coates AS, Gelber RD, Thurlimann B, Senn HJ, et al. Strategies for subtypes: dealing with the diversity of breast cancer: highlights of the St. Gallen International Expert Consensus on the Primary Therapy of Early Breast Cancer 2011. Ann Oncol 2011;22:1736–1747.

56. Guo R, Yu Y, Huang Y, Lin M, Liao Y, Hu Y, et al. A nomogram model combining ultrasound-based radiomics features and clinicopathological factors to identify germline BRCA1/2 mutation in invasive breast cancer patients. Heliyon 2024;10e23383.

57. Cui H, Sun Y, Zhao D, Zhang X, Kong H, Hu N, et al. Radiogenomic analysis of prediction HER2 status in breast cancer by linking ultrasound radiomic feature module with biological functions. J Transl Med 2023;21:44.

58. Huang Y, Qiang Y, Jian L, Jin Z, Lang Q, Sheng C, et al. Ultrasonic features and molecular subtype predict somatic mutations in TP53 and PIK3CA genes in breast cancer. Acad Radiol 2022;29:e261–e270.

59. Huang Y, Guo Y, Xiao Q, Liang S, Yu Q, Qian L, et al. Unraveling the pivotal network of ultrasound and somatic mutations in triple-negative and non-triple-negative breast cancer. Breast Cancer (Dove Med Press) 2023;15:461–472.

60. Park AY, Han MR, Park KH, Kim JS, Son GS, Lee HY, et al. Radiogenomic analysis of breast cancer by using B-mode and vascular US and RNA sequencing. Radiology 2020;295:24–34.

61. Han MR, Park AY, Seo BK, Bae MS, Kim JS, Son GS, et al. Association between vascular ultrasound features and DNA sequencing in breast cancer: a preliminary study. Discov Oncol 2023;14:52.

62. Ma M, Liu R, Wen C, Xu W, Xu Z, Wang S, et al. Predicting the molecular subtype of breast cancer and identifying interpretable imaging features using machine learning algorithms. Eur Radiol 2022;32:1652–1662.

63. Kuchenbaecker KB, Hopper JL, Barnes DR, Phillips KA, Mooij TM, Roos-Blom MJ, et al. Risks of breast, ovarian, and contralateral breast cancer for BRCA1 and BRCA2 mutation carriers. JAMA 2017;317:2402–2416.

64. Lee A, Moon BI, Kim TH. BRCA1/BRCA2 pathogenic variant breast cancer: treatment and prevention strategies. Ann Lab Med 2020;40:114–121.

65. Mao B, Ma J, Duan S, Xia Y, Tao Y, Zhang L. Preoperative classification of primary and metastatic liver cancer via machine learning-based ultrasound radiomics. Eur Radiol 2021;31:4576–4586.

66. Qin X, Hu X, Xiao W, Zhu C, Ma Q, Zhang C. Preoperative evaluation of hepatocellular carcinoma differentiation using contrast-enhanced ultrasound-based deep-learning radiomics model. J Hepatocell Carcinoma 2023;10:157–168.

67. Zheng R, Zhang X, Liu B, Zhang Y, Shen H, Xie X, et al. Comparison of non-radiomics imaging features and radiomics models based on contrast-enhanced ultrasound and Gd-EOB-DTPA-enhanced MRI for predicting microvascular invasion in hepatocellular carcinoma within 5 cm. Eur Radiol 2023;33:6462–6472.

68. Zhang L, Duan S, Qi Q, Li Q, Ren S, Liu S, et al. Noninvasive prediction of Ki-67 expression in hepatocellular carcinoma using machine learning-based ultrasomics: a multicenter study. J Ultrasound Med 2023;42:1113–1122.

69. Qian H, Shen Z, Zhou D, Huang Y. Intratumoral and peritumoral radiomics model based on abdominal ultrasound for predicting Ki-67 expression in patients with hepatocellular cancer. Front Oncol 2023;13:1209111.

70. Wang Y, Weng W, Liang R, Zhou Q, Hu H, Li M, et al. Predicting T cell-inflamed gene expression profile in hepatocellular carcinoma based on dynamic contrast-enhanced ultrasound radiomics. J Hepatocell Carcinoma 2023;10:2291–2303.

71. Bergengren O, Pekala KR, Matsoukas K, Fainberg J, Mungovan SF, Bratt O, et al. 2022 Update on prostate cancer epidemiology and risk factors: a systematic review. Eur Urol 2023;84:191–206.

72. Xiao T, Shen W, Wang Q, Wu G, Yu J, Cui L. The detection of prostate cancer based on ultrasound RF signal. Front Oncol 2022;12:946965.

73. Sun YK, Zhou BY, Miao Y, Shi YL, Xu SH, Wu DM, et al. Three-dimensional convolutional neural network model to identify clinically significant prostate cancer in transrectal ultrasound videos: a prospective, multi-institutional, diagnostic study. EClinicalMedicine 2023;60:102027.

74. Wildeboer RR, Mannaerts CK, van Sloun RJ, Budaus L, Tilki D, Wijkstra H, et al. Automated multiparametric localization of prostate cancer based on B-mode, shear-wave elastography, and contrast-enhanced ultrasound radiomics. Eur Radiol 2020;30:806–815.

75. Wang Y, Feng G, Wang J, An P, Duan P, Hu Y, et al. Contrast-enhanced ultrasound-magnetic resonance imaging radiomics based model for predicting the biochemical recurrence of prostate cancer: a feasibility study. Comput Math Methods Med 2022;2022:8090529.

76. Ageeli W, Zhang X, Ogbonnaya CN, Bray SE, Kernohan NM, Wilson J, et al. Multimodality characterization of cancer-associated fibroblasts in tumor microenvironment and its correlation with ultrasound shear wave-measured tissue stiffness in localized prostate cancer. Front Oncol 2022;12:822476.

77. Fu Q, Luo L, Hong R, Zhou H, Xu X, Feng Y, et al. Radiogenomic analysis of ultrasound phenotypic features coupled to proteomes predicts metastatic risk in primary prostate cancer. BMC Cancer 2024;24:290.

78. Nero C, Ciccarone F, Boldrini L, Lenkowicz J, Paris I, Capoluongo ED, et al. Germline BRCA 1-2 status prediction through ovarian ultrasound images radiogenomics: a hypothesis generating study (PROBE study). Sci Rep 2020;10:16511.

79. Xie X, Shen C, Zhang X, Wu G, Yang B, Qi Z, et al. Rapid intraoperative multi-molecular diagnosis of glioma with ultrasound radio frequency signals and deep learning. EBioMedicine 2023;98:104899.

80. Siegel RL, Miller KD, Fuchs HE, Jemal A. Cancer statistics, 2022. CA Cancer J Clin 2022;72:7–33.

81. Di Pilla A, Nero C, Specchia ML, Ciccarone F, Boldrini L, Lenkowicz J, et al. A cost-effectiveness analysis of an integrated clinical-radiogenomic screening program for the identification of BRCA 1/2 carriers (e-PROBE study). Sci Rep 2024;14:928.

82. Miller KD, Ostrom QT, Kruchko C, Patil N, Tihan T, Cioffi G, et al. Brain and other central nervous system tumor statistics, 2021. CA Cancer J Clin 2021;71:381–406.

83. Louis DN, Perry A, Wesseling P, Brat DJ, Cree IA, Figarella-Branger D, et al. The 2021 WHO classification of tumors of the central nervous system: a summary. Neuro Oncol 2021;23:1231–1251.

84. Jena B, Saxena S, Nayak GK, Balestrieri A, Gupta N, Khanna NN, et al. Brain tumor characterization using radiogenomics in artificial intelligence framework. Cancers (Basel) 2022;14:4052.

85. Gerardi RM, Cannella R, Bonosi L, Vernuccio F, Ferini G, Viola A, et al. Forecasting molecular features in IDH-wildtype gliomas: the state of the art of radiomics applied to neurosurgery. Cancers (Basel) 2023;15:940.

86. Shui L, Ren H, Yang X, Li J, Chen Z, Yi C, et al. The era of radiogenomics in precision medicine: an emerging approach to support diagnosis, treatment decisions, and prognostication in oncology. Front Oncol 2020;10:570465.

87. Ferro M, de Cobelli O, Vartolomei MD, Lucarelli G, Crocetto F, Barone B, et al. Prostate cancer radiogenomics: from imaging to molecular characterization. Int J Mol Sci 2021;22:9971.

88. Singh G, Manjila S, Sakla N, True A, Wardeh AH, Beig N, et al. Radiomics and radiogenomics in gliomas: a contemporary update. Br J Cancer 2021;125:641–657.

89. Zanfardino M, Pane K, Mirabelli P, Salvatore M, Franzese M. TCGA-TCIA impact on radiogenomics cancer research: a systematic review. Int J Mol Sci 2019;20:6033.

90. Gao J, Ye F, Han F, Jiang H, Zhang J. A radiogenomics biomarker based on immunological heterogeneity for non-invasive prognosis of renal clear cell carcinoma. Front Immunol 2022;13:956679.

91. Liu S, Wang Y, Yang X, Lei B, Liu L, Li SX, et al. Deep learning in medical ultrasound analysis: a review. Engineering 2019;5:261–275.

92. Wang LF, Wang Q, Mao F, Xu SH, Sun LP, Wu TF, et al. Risk stratification of gallbladder masses by machine learning-based ultrasound radiomics models: a prospective and multi-institutional study. Eur Radiol 2023;33:8899–8911.

93. Yang Y, Chen F, Liang H, Bai Y, Wang Z, Zhao L, et al. CNN-based automatic segmentations and radiomics feature reliability on contrast-enhanced ultrasound images for renal tumors. Front Oncol 2023;13:1166988.

94. Chen LD, Li W, Xian MF, Zheng X, Lin Y, Liu BX, et al. Preoperative prediction of tumour deposits in rectal cancer by an artificial neural network-based US radiomics model. Eur Radiol 2020;30:1969–1979.

95. Moro F, Albanese M, Boldrini L, Chiappa V, Lenkowicz J, Bertolina F, et al. Developing and validating ultrasound-based radiomics models for predicting high-risk endometrial cancer. Ultrasound Obstet Gynecol 2022;60:256–268.

96. Hu Y, Li A, Zhao CK, Ye XH, Peng XJ, Wang PP, et al. A multiparametric clinic-ultrasomics nomogram for predicting extremity soft-tissue tumor malignancy: a combined retrospective and prospective bicentric study. Radiol Med 2023;128:784–797.

Article information Continued

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/licenses/by-nc/4.0/) which permits unrestricted noncommercial use, distribution, and reproduction in any medium, provided the original work is properly cited.

Notes

Key points

Radiogenomics, when applied to ultrasound (US), examines the relationships between US imaging features and gene expression patterns. The implementation of US-based radiogenomics offers a rapid and non-invasive approach to identifying gene variations and molecular subtypes in cancers. Future US-based radiogenomics research should prioritize prospective, multicenter studies with large sample sizes and standardization.

Type	Reference	Molecule studied	Patients	Imaging modality	Segmentation	Method	Results
Thyroid	Yoon et al. [26]	BRAF mutation	527 (training: validation, 387:140)	BUS	Manual	Lasso regression	The radiomics score demonstrated good discriminatory power for BRAF V600E mutation of PTCs in the training group (AUC, 0.718; 95% CI, 0.650 to 0.786) but a low AUC of 0.692 (95% CI, 0.516 to 0.742) in the validation group.
	Kwon et al. [27]	BRAF mutation	96 (training: validation, 77: 18)	BUS	Manual	Logistic regression, SVM, random forest	The radiomics approach revealed that three classification models demonstrated moderate performance in predicting BRAF mutation in PTCs, achieving an AUC of 0.651 and an accuracy of 64.3% on average.
	Wang et al. [28]	BRAF mutation	138 (training: validation, 96:42)	BUS, UE	Manual	Lasso regression	The radiomics model using BUS and UE images exhibited high predictive ability (AUC, 0.938; 95% CI, 0.851 to 1.000) for BRAF V600E mutation in patients with PTC.
	Agyekum et al. [29]	BRAF mutation	138 (training: validation, 96:42)	UE	Manual	SVM_L, SVM_RBF, logistic regression, NB, KNN, LDA	SVM_RBF could predict BRAF V600E mutation in PTC, achieving the best AUC (0.98), ACC (0.93), SEN (0.95), SPEC (0.90), PPV (0.91), and NPV (0.95).
	Yu et al. [30]	RET rearrangement	630	BUS	Manual, CNN	SVM, KNN, logistic regression, random forest, decision tree, extra trees, LightGBM, XGBoost, ResNet50	The deep learning radiomics nomogram achieved an AUC of 0.9545 (95% CI, 0.9133 to 0.9558) in the test set and an AUC of 0.9396 (95% CI, 0.9185 to 0.9607) in the training set.
	Tong et al. [31]	Gene expression	270 (training: validation, 180:90)	BUS	Manual	SVM	Nine significant relationships were found between radiomics characteristics and gene modules in the radiogenomic map, with two of these demonstrating particularly high correlation coefficients.
	Lee et al. [32]	Differentially expressed genes	273	BUS, CDFI	Manual	CNN (ResNet50, ResNet101, VGG16)	Radiologists and CNNs identified specific US characteristics that were indicative of upregulation in genes associated with immune response and epithelial-mesenchymal transition.

Reference	Molecule studied	Patients	Imaging modality	Segmentation	Method	Results
Guo et al. [44]	ER/PR/HER2	215	BUS	Semi-automatic	SVM	The radiomics method revealed a significant association between receptor status and US features (AUC, 0.760; 95% CI, 0.755 to 0.764).
Xu et al. [45]	ER, PR, HER2, and Ki-67	342	BUS	Manual	SVM	In the validation set, the AUCs for predicting ER, PR, HER2, and Ki-67 expression were 0.868, 0.811, 0.722, and 0.706, respectively.
Xu et al. [46]	ER, PR, HER2, and Ki-67	489 (training: validation ratio, 317:137)	BUS	Manual	Logistic regression	A radiomics nomogram model had an AUC of 0.837 and 0.813 in the training dataset and validation datasets, respectively.
Jiang et al. [47]	ER, PR, HER2, and Ki-67	1,275	BUS	CNN	CNN	In the two test cohorts, a deep CNN distinguished luminal disease from non-luminal breast cancer with AUCs of 0.87 (95% CI, 0.83 to 0.91) and 0.83 (95% CI, 0.78 to 0.87), respectively.
Zhou et al. [48]	ER, PR, HER2, and Ki-67	818 (training: validation, 545:263)	BUS, CDFI, SWE	CNN	CNN (DenseNet 121, ResNet 50, SENet 50)	A multimodal ACNN model demonstrated the best performance for the prediction of four- classification molecular subtypes (AUC, 0.89 to 0.96) and five-classification molecular subtypes (AUC, 0.87 to 0.94) in breast cancer.
Quan et al. [49]	HER2	445 (training: validation, 356:89)	BUS (video)	CNN	SVM, random forest, logistic regression, XGBoost	A deep learning radiomics model using XGBoost and logistic regression classifiers performed best in predicting HER2 expression status, with a specificity of 0.917 and an AUC of 0.81 for the test group.
Gong et al. [50]	ER, PR, HER2, and Ki-67	119	BUS, CEUS	Manual	Multivariate logistic regression	CEUS video improved the performance of the standard US radiomics model in the prediction of Luminal A, HER2 overexpression, HR positivity, and HER2 positivity in breast cancer (ACC, 70.2%, 84.0%, 74.5%, and 72.5%; P<0.01).
Ma et al. [51]	ER, PR, HER2, and Ki-67	600 (training: validation, 450:150)	BUS, mammography	-	Decision tree, KNN, logistic regression, NB, random forest	A decision tree model outperformed the other models in distinguishing TNBC from other subtypes of breast cancer (AUC, 0.971).

Reference	Molecule Studied	Patients	Imaging modality	Segmentation	Method	Results
Guo et al. [56]	BRCA1/2 mutation	449 (490 lesions, training: validation ratio, 7:3)	BUS	Manual	Multivariate logistic regression	A nomogram model that combined radiomics and clinical features exhibited high AUCs: 0.804 (95% CI, 0.748 to 0.861) in the training set and 0.811 (95% CI, 0.724 to 0.894) in the validation set.
Deng et al. [51]	BRCA1/2 mutation	497 (training: validation, 348:149)	BUS	Manual	Lasso regression, logistic regression	A nomogram developed from radiomics signatures and clinicopathologic predictors achieved AUCs of 0.850 (95% CI, 0.803 to 0.898) and 0.824 (95% CI, 0.755 to 0.894) in the training and validation sets, respectively.
Cui et al. [57]	HER2, mRNA expression profiles	489 (training: validation ratio, 7:3)	BUS	CNN	SVM, random forest, decision tree, logistic regression, NB, ANN, KNN	A radiomics model derived from differential US radiomics features demonstrated discriminative ability in HER2 status prediction (AUC, 0.844; 95% CI, 0.762 to 0.927).
Huang et al. [58]	TP53, PIK3CA	386 (training: validation, 243:143)	BUS, CDFI	-	Multivariate logistic regression	In the validation set, the multivariate model exhibited an AUC of 0.715 for predicting PIK3CA mutation and an AUC of 0.653 for predicting TP53 mutation. Distinct US radiomics features were
Huang et al. [59]	Sequencing of 511 genes	166	BUS	Automatic	Elastic net regression, logistic regression	linked to different biological processes in patients with and without TNBC.
Park et al. [60]	RNA sequencing	31	BUS, CDFI, CEUS	-	-	Thirteen US phenotypes were linked to various patterns of 340 differentially expressed genes (P<0.05).
Han et al. [61]	Sequencing of 105 genes	31	CDFI, CEUS	-	-	A significant association was found between eight ultrasonography characteristics and nine single nucleotide polymorphisms (P<0.05).

Type	Reference	Molecule studied	Patients	Imaging modality	Segmentation	Method	Results
Liver	Wang et al. [70]	RNA sequencing	268	CEUS	Manual	Logistic regression	A model developed based on CEUS arterial phase features performed well in predicting the T cell- inflamed gene expression profile, with a mean AUC of 0.905.
Prostate	Ageeli et al. [76]	RNA sequencing	30	SWE	-	-	Tissue stiffness in the tumor stroma was positively correlated with gene expression of SMAα and PDGFRβ in the fibromuscular stroma (P<0.001).
Ovarian	Nero et al. [78]	BRCA1/2	255 (training: validation ratio, 3:1)	BUS	Manual	Logistic regression, SVM, decision tree, automated machine learning pipelines	The four strategies performed similarly in predicting gBRCA1/2 status, with ACC ranging from 0.54 for logistic regression to 0.64 for the automated machine learning pipeline in the testing set.
Brain	Xie et al. [79]	IDH1, TERTp, 1p/19q	103	US RF	-	Deep learning	A spatial-temporal integration model demonstrated a mean accuracy of 0.9190 and a mean AUC of 0.9650 (95% CI, 0.94 to 0.99) for diagnosing molecular biomarkers in the retrospective cohort, as well as an average accuracy exceeding 80% in clinical testing.