Multi-marker quantitative radiomics for mass characterization in dedicated breast CT imaging

Gradient-Based Radiomics for Outcome Prediction and Decision-Making in PULSAR: A Preliminary Study

Purpose

Personalized ultrafractionated stereotactic adaptive radiation therapy (PULSAR) has emerged as an innovative method for delivering high-dose radiation over extended intervals, adapting treatment based on the patient's response. However, current adaptation largely relies on physicians' experience and tumor size assessment, underscoring the need for a data-driven approach to improve outcome prediction and support decision-making.

Materials and Methods

We analyzed 69 lesions from 39 patients undergoing PULSAR treatment. Gradient-based features, including gradient magnitude, radial gradient, and radial deviation, were extracted from both intratumoral and peritumoral regions, with the latter further divided into octant subregions. Support vector machine models were developed using features from first magnetic resonance imaging (MRI), second MRI, and delta mode (change between the 2). An ensemble feature selection (EFS) model was then created by combining the features of the top-performing individual models. The approach was validated on a non-PULSAR cohort (37 lesions from 23 patients) treated with standard fractionated stereotactic radiation therapy.

Results

The EFS model shows strong predictive performance in determining whether tumor volume reduction exceeds 20% at the 3-month postradiation time point. Features derived from octant subregions exhibit significantly better prediction than those from the core or entire margin. Pretreatment features (from first MRI) generally outperform second MRI and delta-mode features, while the inclusion of 1 delta feature in the EFS model enhances performance. In the non-PULSAR cohort, the gradient-based approach outperforms conventional radiomics, demonstrating its strong generalizability.

Conclusion

Our gradient-based radiomics approach, combining spatial segmentation and temporal features, significantly enhances treatment response prediction in PULSAR therapy. Its superior performance compared to conventional radiomics, coupled with its effectiveness in both PULSAR and non-PULSAR cohorts, highlights its potential as a robust tool for personalized treatment planning in neuro-oncology, applicable to both photon and particle therapies.

Cone-beam Breast CT Features Associated With Lymphovascular Invasion in Patients With Breast Cancer

RATIONALE AND OBJECTIVES

To explore the association between contrast-enhanced cone-beam breast CT (CE-CBBCT) imaging features and lymphovascular invasion (LVI) status.

MATERIALS AND METHODS

Female patients with breast cancer by postoperative histopathology who underwent preoperative CE-CBBCT from August 2020 to December 2023 were retrospectively enrolled. Two radiologists reviewed the CBBCT image features independently. Clinicopathologic and CBBCT characteristics between LVI-positive and LVI-negative were compared by χ2 or Fisher's exact tests and Student's t or Mann-Whitney U test, as appropriate. Multivariate logistic regression analysis was performed to identify independent predictive factors of LVI. The receiver operating characteristic curve was used to evaluate predictive performance.

RESULTS

A total of 401 women were enrolled. LVI status of breast cancer was significantly associated with histologic type, Ki-67 index, adjacent vessel sign (AVS), increased ipsilateral whole-breast vascularity (IIV) number, and IIV degree (all p<0.05). In mass, calcification, AVS, IIV number, and IIV degree were significantly associated with LVI (all p<0.05). In non-mass enhancement (NME), AVS, IIV number, and IIV degree were associated with LVI (all p<0.05). Multivariate logistic regression showed AVS (OR=4.367, p<0.001) and IIV degree (OR of moderate and prominent IIV=4.732, 3.641, both p<0.005) as independent risk factors for LVI. Specifically, in mass, AVS (OR=4.397, p<0.001) and moderate-to-prominent IIV (OR=4.815, 3.563, both p<0.01) were independent predictors. For NME, moderate-to-prominent IIV (OR=13.695, 4.054, both p=0.001) was also an independent factor. The combined LVI prediction model which included AVS and IIV degree showed excellent performance (AUC=0.804).

CONCLUSION

CBBCT imaging features can help identify LVI status in breast cancer patients, which will guide the accurate planning of treatment management.

Multiomics-Based Outcome Prediction in Personalized Ultra-Fractionated Stereotactic Adaptive Radiotherapy (PULSAR)

Objectives: This retrospective study aims to develop a multiomics approach that integrates radiomics, dosiomics, and delta features to predict treatment responses in brain metastasis (BM) patients undergoing PULSAR. Methods: A retrospective study encompassing 39 BM patients with 69 lesions treated with PULSAR was undertaken. Radiomics, dosiomics, and delta features were extracted from both pre-treatment and intra-treatment MRI scans alongside dose distributions. Six individual models, alongside an ensemble feature selection (EFS) model, were evaluated. The classification task focused on distinguishing between two lesion groups based on whether they exhibited a volume reduction of more than 20% at follow-up. Performance metrics, including sensitivity, specificity, accuracy, precision, F1 score, and the area under the receiver operating characteristic (ROC) curve (AUC), were assessed. Results: The EFS model integrated the features from pre-treatment radiomics, pre-treatment dosiomics, intra-treatment radiomics, and delta radiomics. It outperformed six individual models, achieving an AUC of 0.979, accuracy of 0.917, and F1 score of 0.821. Among the top nine features of the EFS model, six features came from post-wavelet transformation and three from original images. Conclusions: The study demonstrated the feasibility of employing a data-driven multiomics approach to predict treatment outcomes in BM patients receiving PULSAR treatment. Integrating multiomics with intra-treatment decision support in PULSAR shows promise for optimizing patient management and reducing the risks of under- or over-treatment.

Radiomics in cone-beam breast CT for the prediction of axillary lymph node metastasis in breast cancer: a multi-center multi-device study

OBJECTIVES

To develop a radiomics model in contrast-enhanced cone-beam breast CT (CE-CBBCT) for preoperative prediction of axillary lymph node (ALN) status and metastatic burden of breast cancer.

METHODS

Two hundred and seventy-four patients who underwent CE-CBBCT examination with two scanners between 2012 and 2021 from two institutions were enrolled. The primary tumor was annotated in each patient image, from which 1781 radiomics features were extracted with PyRadiomics. After feature selection, support vector machine models were developed to predict ALN status and metastatic burden. To avoid overfitting on a specific patient subset, 100 randomly stratified splits were made to assign the patients to either training/fine-tuning or test set. Area under the receiver operating characteristic curve (AUC) of these radiomics models was compared to those obtained when training the models only with clinical features and combined clinical-radiomics descriptors. Ground truth was established by histopathology.

RESULTS

One hundred and eighteen patients had ALN metastasis (N + (≥ 1)). Of these, 74 had low burden (N + (1~2)) and 44 high burden (N + (≥ 3)). The remaining 156 patients had none (N0). AUC values across the 100 test repeats in predicting ALN status (N0/N + (≥ 1)) were 0.75 ± 0.05 (0.67~0.93, radiomics model), 0.68 ± 0.07 (0.53~0.85, clinical model), and 0.74 ± 0.05 (0.67~0.88, combined model). For metastatic burden prediction (N + (1~2)/N + (≥ 3)), AUC values were 0.65 ± 0.10 (0.50~0.88, radiomics model), 0.55 ± 0.10 (0.40~0.80, clinical model), and 0.64 ± 0.09 (0.50~0.90, combined model), with all the ranges spanning 0.5. In both cases, the radiomics model was significantly better than the clinical model (both p < 0.01) and comparable with the combined model (p = 0.56 and 0.64).

CONCLUSIONS

Radiomics features of primary tumors could have potential in predicting ALN metastasis in CE-CBBCT imaging.

CLINICAL RELEVANCE STATEMENT

The findings support potential clinical use of radiomics for predicting axillary lymph node metastasis in breast cancer patients and addressing the limited axilla coverage of cone-beam breast CT.

KEY POINTS

• Contrast-enhanced cone-beam breast CT-based radiomics could have potential to predict N0 vs. N + (≥ 1) and, to a limited extent, N + (1~2) vs. N + (≥ 3) from primary tumor, and this could help address the limited axilla coverage, pending future verifications on larger cohorts. • The average AUC of radiomics and combined models was significantly higher than that of clinical models but showed no significant difference between themselves. • Radiomics features descriptive of tumor texture were found informative on axillary lymph node status, highlighting a higher heterogeneity for tumor with positive axillary lymph node.

Potential added value of computed tomography radiomics to multimodal prediction models for benign and malignant breast tumors

Background

Early diagnosis is crucial to the treatment of breast cancer, but conventional imaging detection is challenging. Radiomics has the potential to improve early diagnostic efficacy in a noninvasive manner. This study examined whether integrating computed tomography (CT) radiomics information based on ultrasound (US) models can improve the efficacy of breast cancer prediction.

Methods

We retrospectively analyzed 420 patients with pathologically confirmed benign or malignant breast tumors. Clinical data and examination images were collected, and the population was divided into training (n=294) and validation (n=126) groups at a ratio of 7:3. The region of interest (ROI) was manually segmented along the tumor boundary using MaZda software, and the features of each ROI was extracted. After dimension reduction and screening, the best features were retained. Subsequently, random forest (RF), support vector machines, and K-nearest neighbor classifiers were used to establish prediction models in an US and combined-methods group.

Results

Finally, 8 of the 379 features were retained in the US group. Random forest was found to be the best model, and the area under the curve (AUC) of the training and validation groups was 0.90 [95% confidence interval (CI): 0.852-0.942] and 0.85 (95% CI: 0.775-0.930), respectively. Meanwhile, 12 of the 750 features were retained in the combined group. In this regard, random forest proved to be the best model, and the AUC of the training and validation group was 0.95 (95% CI: 0.918-0.981) and 0.92 (95% CI: 0.866-0.969), respectively. The calibration curve showed a good fit of the model. The decision curve showed that the clinical net benefit of the combined group was far greater than that of any single examination, and the prediction model of the combined group exhibited a degree of practical clinical value.

Conclusions

The combined model based on US and CT images has potential application value in the prognostic prediction of benign and malignant breast diseases.

Radiomics nomogram for predicting axillary lymph node metastasis-a potential method to address the limitation of axilla coverage in cone-beam breast CT: a bi-center retrospective study

PURPOSE

Cone-beam breast CT (CBBCT) has an inherent limitation that the axilla cannot be imaged in its entirety. We aimed to develop and validate a nomogram based on clinical factors and contrast-enhanced (CE) CBBCT radiomics features to predict axillary lymph node (ALN) metastasis and complement limited axilla coverage.

MATERIAL AND METHODS

This retrospective study included 312 patients with breast cancer from two hospitals who underwent CE-CBBCT examination in a clinical trial (NCT01792999) during 2012-2020. Patients from TCIH comprised training set (n = 176) and validation set (n = 43), and patients from SYSUCC comprised external test set (n = 93). 3D ROIs were delineated manually and radiomics features were extracted by 3D Slicer software. RadScore was calculated and radiomics model was constructed after feature selection. Clinical model was built on independent predictors. Nomogram was developed with independent clinical predictors and RadScore. Diagnostic performance was compared among three models by ROC curve, and decision curve analysis (DCA) was used to evaluate the clinical utility of nomogram.

RESULTS

A total of 139 patients were ALN positive and 173 patients were negative. Twelve radiomics features remained after feature selection. Location and focality were selected as independent predictors for ALN status. The AUC of nomogram in external test set was higher than that of clinical model (0.80 vs. 0.66, p = 0.012). DCA demonstrated that the nomogram had higher overall net benefit than that of clinical model.

CONCLUSION

The nomogram combined CE-CBBCT-based radiomics features and clinical factors could have potential in distinguishing ALN positive from negative and addressing the limitation of axilla coverage in CBBCT.

Radiomics to Differentiate Malignant and Benign Breast Lesions: A Systematic Review and Diagnostic Test Accuracy Meta-Analysis

Breast cancer is a prevalent global health concern, necessitating accurate diagnostic tools for effective management. Diagnostic imaging plays a pivotal role in breast cancer diagnosis, staging, treatment planning, and outcome evaluation. Radiomics is an emerging field of study in medical imaging that contains a broad set of computational methods to extract quantitative features from radiographic images. This can be utilized to guide diagnosis, treatment response, and prognosis in clinical settings.  A systematic review was performed in concordance with Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines and the Cochrane Handbook for Systematic Reviews of Diagnostic Test Accuracy. Quality was assessed using the radiomics quality score. Diagnostic sensitivity and specificity of radiomics analysis, with 95% confidence intervals (CIs), were included for meta-analysis. The area under the curve analysis was recorded. An extensive statistical analysis was performed following the Cochrane guidelines. Statistical significance was determined if p-values were less than 0.05. Statistical analyses were conducted using Review Manager (RevMan), Version 5.4.1. A total of 31 manuscripts involving 8,773 patients were included, with 17 contributing to the meta-analysis. The cohort comprised 56.2% malignant breast cancers and 43.8% benign breast lesions. MRI demonstrated a sensitivity of 0.91 (95% CI: 0.89-0.92) and a specificity of 0.84 (95% CI: 0.82-0.86) in differentiating between benign and malignant breast cancers. Mammography-based radiomic features predicted breast cancer subtype with a sensitivity of 0.79 (95% CI: 0.76-0.82) and a specificity of 0.81 (95% CI: 0.79-0.84). Ultrasound-based analysis yielded a sensitivity of 0.92 (95% CI: 0.90-0.94) and a specificity of 0.85 (95% CI: 0.83-0.88). Only one study reported the results of radiomic evaluation from CT, which had a sensitivity of 0.95 (95% CI: 0.88-0.99) and a specificity of 0.56 (95% CI: 0.45-0.67).  Across different imaging modalities, radiomics exhibited robust diagnostic accuracy in differentiating benign and malignant breast lesions. The results underscore the potential of radiomic assessment as a minimally invasive alternative or adjunctive diagnostic tool for breast cancer. This is pioneering data that reports on a novel diagnostic approach that is understudied and underreported. However, due to study limitations, the complexity of this technology, and the need for future development, biopsy still remains the current gold standard method of determining breast cancer type.

Diagnostic accuracy of cone-beam breast computed tomography and head-to-head comparison of digital mammography, magnetic resonance imaging and cone-beam breast computed tomography for breast cancer: a systematic review and meta-analysis

Background

Cone-beam breast computed tomography (CBBCT) is a new breast imaging technique, however, CBBCT is not yet widely used, and its future application will depend on its diagnostic potential and application value. Therefore, it is of great clinical significance to systematically review and analyze the diagnostic accuracy of CBBCT for breast cancer detection in existing studies and compare it with other traditional imaging methods for the diagnosis of breast lesions.

Methods

We searched PubMed, Embase, Web of Science, and Chinese databases until August 2022 for relevant papers. Studies evaluating the diagnostic accuracy of CBBCT in women with suspected breast cancer were included. Each study's quality was evaluated using the Quality Assessment of Diagnostic Performance Studies-2 (QUADAS-2) instrument.

Results

Eighteen studies with a total of 1,792 patients were included in the analysis. The overall pooled sensitivity and specificity of CBBCT in diagnosing breast cancer were 0.95 [95% confidence interval (CI): 0.91-0.97] and 0.72 (95% CI: 0.62-0.80), respectively. The area under the curve (AUC) for CBBCT was 0.92 (95% CI: 0.90-0.94). In a head-to-head comparison of CBBCT and digital mammography (DM), eight trials with 992 patients were included in the study, and the AUCs for CBBCT and DM were 0.94 (95% CI: 0.92-0.96) and 0.83 (95% CI: 0.80-0.83), respectively. In a head-to-head comparison of CBBCT and magnetic resonance imaging (MRI), four trials with 203 patients were included in the analysis; the AUC for CBBCT and MRI were 0.88 (95% CI: 0.85-0.91) and 0.96 (95% CI: 0.94-0.97), respectively.

Conclusions

This meta-analysis of CBBCT test accuracy indicated encouraging diagnostic performance. In the summary of head-to-head comparative studies, there is a tendency for CBBCT to have greater diagnostic accuracy than DM, although its diagnostic performance is marginally inferior to that of MRI. However, the meta-analysis results were derived from studies with limited sample sizes. There is a need for more extensive research in this setting.

AI-Based Cancer Detection Model for Contrast-Enhanced Mammography

BACKGROUND

The recent development of deep neural network models for the analysis of breast images has been a breakthrough in computer-aided diagnostics (CAD). Contrast-enhanced mammography (CEM) is a recent mammography modality providing anatomical and functional imaging of the breast. Despite the clinical benefits it could bring, only a few research studies have been conducted around deep-learning (DL) based CAD for CEM, especially because the access to large databases is still limited. This study presents the development and evaluation of a CEM-CAD for enhancing lesion detection and breast classification.

MATERIALS & METHODS

A deep learning enhanced cancer detection model based on a YOLO architecture has been optimized and trained on a large CEM dataset of 1673 patients (7443 images) with biopsy-proven lesions from various hospitals and acquisition systems. The evaluation was conducted using metrics derived from the free receiver operating characteristic (FROC) for the lesion detection and the receiver operating characteristic (ROC) to evaluate the overall breast classification performance. The performances were evaluated for different types of image input and for each patient background parenchymal enhancement (BPE) level.

RESULTS

The optimized model achieved an area under the curve (AUROC) of 0.964 for breast classification. Using both low-energy and recombined image as inputs for the DL model shows greater performance than using only the recombined image. For the lesion detection, the model was able to detect 90% of all cancers with a false positive (non-cancer) rate of 0.128 per image. This study demonstrates a high impact of BPE on classification and detection performance.

CONCLUSION

The developed CEM CAD outperforms previously published papers and its performance is comparable to radiologist-reported classification and detection capability.

Four-Dimensional Machine Learning Radiomics for the Pretreatment Assessment of Breast Cancer Pathologic Complete Response to Neoadjuvant Chemotherapy in Dynamic Contrast-Enhanced MRI

BACKGROUND

Breast cancer response to neoadjuvant chemotherapy (NAC) is typically evaluated through the assessment of tumor size reduction after a few cycles of NAC. In case of treatment ineffectiveness, this results in the patient suffering potentially severe secondary effects without achieving any actual benefit.

PURPOSE

To identify patients achieving pathologic complete response (pCR) after NAC by spatio-temporal radiomic analysis of dynamic contrast-enhanced (DCE) MRI images acquired before treatment.

STUDY TYPE

Single-center, retrospective.

POPULATION

A total of 251 DCE-MRI pretreatment images of breast cancer patients.

FIELD STRENGTH/SEQUENCE

1.5 T/3 T, T1-weighted DCE-MRI.

ASSESSMENT

Tumor and peritumoral regions were segmented, and 348 radiomic features that quantify texture temporal variation, enhancement kinetics heterogeneity, and morphology were extracted. Based on subsets of features identified through forward selection, machine learning (ML) logistic regression models were trained separately with all images and stratifying on cancer molecular subtype and validated with leave-one-out cross-validation.

STATISTICAL TESTS

Feature significance was assessed using the Mann-Whitney U-test. Significance of the area under the receiver operating characteristics (ROC) curve (AUC) of the ML models was assessed using the associated 95% confidence interval (CI). Significance threshold was set to 0.05, adjusted with Bonferroni correction.

RESULTS

Nine features related to texture temporal variation and enhancement kinetics heterogeneity were significant in the discrimination of cases achieving pCR vs. non-pCR. The ML models achieved significant AUC of 0.707 (all cancers, n = 251, 59 pCR), 0.824 (luminal A, n = 107, 14 pCR), 0.823 (luminal B, n = 47, 15 pCR), 0.844 (HER2 enriched, n = 25, 11 pCR), 0.803 (triple negative, n = 72, 19 pCR).

DATA CONCLUSIONS

Differences in imaging phenotypes were found between complete and noncomplete responders. Furthermore, ML models trained per cancer subtype achieved high performance in classifying pCR vs. non-pCR cases. They may, therefore, have potential to help stratify patients according to the level of response predicted before treatment, pending further validation with larger prospective cohorts.

EVIDENCE LEVEL

4 TECHNICAL EFFICACY: Stage 4.

Predictive Value of 18F-FDG PET/CT-Based Radiomics Model for Occult Axillary Lymph Node Metastasis in Clinically Node-Negative Breast Cancer

Background: To develop and validate a radiomics model based on ¹⁸F-FDG PET/CT images to preoperatively predict occult axillary lymph node (ALN) metastases in patients with invasive ductal breast cancer (IDC) with clinically node-negative (cN0); Methods: A total of 180 patients (mean age, 55 years; range, 31–82 years) with pathologically proven IDC and a preoperative ¹⁸F-FDG PET/CT scan from January 2013 to January 2021 were included in this retrospective study. According to the intraoperative pathological results of ALN, we divided patients into the true-negative group and ALN occult metastasis group. Radiomics features were extracted from PET/CT images using Pyradiomics implemented in Python, t-tests, and LASSO were used to screen the feature, and the random forest (RF), support vector machine (SVM), stochastic gradient descent (SGD), and k-nearest neighbor (KNN) were used to build the prediction models. The best-performing model was further tested by the permutation test; Results: Among the four models, RF had the best prediction results, the AUC range of RF was 0.661–0.929 (mean AUC, 0.817), and the accuracy range was 65.3–93.9% (mean accuracy, 81.2%). The p-values of the permutation tests for the RF model with maximum and minimum accuracy were less than 0.01; Conclusions: The developed RF model was able to predict occult ALN metastases in IDC patients based on preoperative ¹⁸F-FDG PET/CT radiomic features.

FMRNet: A fused network of multiple tumoral regions for breast tumor classification with ultrasound images

PURPOSE

Recent studies have illustrated that the peritumoral regions of medical images have value for clinical diagnosis. However, the existing approaches using peritumoral regions mainly focus on the diagnostic capability of the single region and ignore the advantages of effectively fusing the intratumoral and peritumoral regions. In addition, these methods need accurate segmentation masks in the testing stage, which are tedious and inconvenient in clinical applications. To address these issues, we construct a deep convolutional neural network that can adaptively fuse the information of multiple tumoral-regions (FMRNet) for breast tumor classification using ultrasound (US) images without segmentation masks in the testing stage.

METHODS

To sufficiently excavate the potential relationship, we design a fused network and two independent modules to extract and fuse features of multiple regions simultaneously. First, we introduce two enhanced combined-tumoral (EC) region modules, aiming to enhance the combined-tumoral features gradually. Then, we further design a three-branch module for extracting and fusing the features of intratumoral, peritumoral, and combined-tumoral regions, denoted as the intratumoral, peritumoral, and combined-tumoral module. Especially, we design a novel fusion module by introducing a channel attention module to adaptively fuse the features of three regions. The model is evaluated on two public datasets including UDIAT and BUSI with breast tumor ultrasound images. Two independent groups of experiments are performed on two respective datasets using the fivefold stratified cross-validation strategy. Finally, we conduct ablation experiments on two datasets, in which BUSI is used as the training set and UDIAT is used as the testing set.

RESULTS

We conduct detailed ablation experiments about the proposed two modules and comparative experiments with other existing representative methods. The experimental results show that the proposed method yields state-of-the-art performance on both two datasets. Especially, in the UDIAT dataset, the proposed FMRNet achieves a high accuracy of 0.945 and a specificity of 0.945, respectively. Moreover, the precision (PRE = 0.909) even dramatically improves by 21.6% on the BUSI dataset compared with the existing method of the best result.

CONCLUSION

The proposed FMRNet shows good performance in breast tumor classification with US images, and proves its capability of exploiting and fusing the information of multiple tumoral-regions. Furthermore, the FMRNet has potential value in classifying other types of cancers using multiple tumoral-regions of other kinds of medical images.

Dedicated breast CT: state of the art-Part II. Clinical application and future outlook

Dedicated breast CT is being increasingly used for breast imaging. This technique provides images with no compression, removal of tissue overlap, rapid acquisition, and available simultaneous assessment of microcalcifications and contrast enhancement. In this second installment in a 2-part review, the current status of clinical applications and ongoing efforts to develop new imaging systems are discussed, with particular emphasis on how to achieve optimized practice including lesion detection and characterization, response to therapy monitoring, density assessment, intervention, and implant evaluation. The potential for future screening with breast CT is also addressed. KEY POINTS: • Dedicated breast CT is an emerging modality with enormous potential in the future of breast imaging by addressing numerous clinical needs from diagnosis to treatment. • Breast CT shows either noninferiority or superiority with mammography and numerical comparability to MRI after contrast administration in diagnostic statistics, demonstrates excellent performance in lesion characterization, density assessment, and intervention, and exhibits promise in implant evaluation, while potential application to breast cancer screening is still controversial. • New imaging modalities such as phase-contrast breast CT, spectral breast CT, and hybrid imaging are in the progress of R & D.

Computer-aided diagnosis of masses in breast computed tomography imaging: deep learning model with combined handcrafted and convolutional radiomic features

Purpose: A computer-aided diagnosis (CADx) system for breast masses is proposed, which incorporates both handcrafted and convolutional radiomic features embedded into a single deep learning model. Approach: The model combines handcrafted and convolutional radiomic signatures into a multi-view architecture, which retrieves three-dimensional (3D) image information by simultaneously processing multiple two-dimensional mass patches extracted along different planes through the 3D mass volume. Each patch is processed by a stream composed of two concatenated parallel branches: a multi-layer perceptron fed with automatically extracted handcrafted radiomic features, and a convolutional neural network, for which discriminant features are learned from the input patches. All streams are then concatenated together into a final architecture, where all network weights are shared and the learning occurs simultaneously for each stream and branch. The CADx system was developed and tested for diagnosis of breast masses ( N = 284 ) using image datasets acquired with independent dedicated breast computed tomography systems from two different institutions. The diagnostic classification performance of the CADx system was compared against other machine and deep learning architectures adopting handcrafted and convolutional approaches, and three board-certified breast radiologists. Results: On a test set of 82 masses (45 benign, 37 malignant), the proposed CADx system performed better than all other model architectures evaluated, with an increase in the area under the receiver operating characteristics curve (AUC) of 0.05 ± 0.02 , and achieving a final AUC of 0.947, outperforming the three radiologists ( AUC = 0.814 - 0.902 ). Conclusions: In conclusion, the system demonstrated its potential usefulness in breast cancer diagnosis by improving mass malignancy assessment.