Evaluation of the peritumoral features using radiomics and deep learning technology in non-spiculated and noncalcified masses of the breast on mammography

Abstract

Objective: To assess the significance of peritumoral features based on deep learning in classifying non-spiculated and noncalcified masses (NSNCM) on mammography.

Methods: We retrospectively screened the digital mammography data of 2254 patients who underwent surgery for breast lesions in Harbin Medical University Cancer Hospital from January to December 2018. Deep learning and radiomics models were constructed. The classification efficacy in ROI and patient levels of AUC, accuracy, sensitivity, and specificity were compared. Stratified analysis was conducted to analyze the influence of primary factors on the AUC of the deep learning model. The image filter and CAM were used to visualize the radiomics and depth features.

Results: For 1298 included patients, 771 (59.4%) were benign, and 527 (40.6%) were malignant. The best model was the deep learning combined model (2 mm), in which the AUC was 0.884 (P < 0.05); especially the AUC of breast composition B reached 0.941. All the deep learning models were superior to the radiomics models (P < 0.05), and the class activation map (CAM) showed a high expression of signals around the tumor of the deep learning model. The deep learning model achieved higher AUC for large size, age >60 years, and breast composition type B (P < 0.05).

Conclusion: Combining the tumoral and peritumoral features resulted in better identification of malignant NSNCM on mammography, and the performance of the deep learning model exceeded the radiomics model. Age, tumor size, and the breast composition type are essential for diagnosis.

Keywords: deep learning; mammography; non-spiculated and noncalcified masses; peritumoral features; radiomics.

Figure 1

Overall flowchart of the method. Tumor region (red line). Peritumoral regions (lines with different colors outside red line). The radiomics and deep models respectively for benign/malignant prediction.

Figure 2

Flow chart of inclusion and exclusion of patients and division of data set.

Figure 3

Significant radiomics features of masses in tumor ROI. Benign masses (A, B). Malignant masses (C–F).

Figure 4-1

CAM status of tumoral and peritumoral regions were observed on 8 patients with malignant masses. All the regions were correctly predicted.

Figure 4-2

CAM status of tumoral and peritumoral regions were observed on 6 patients with malignant masses. Tumoral region predictions were wrong. Proximal peritumoral region predictions as follows:1) regions of both 1 mm and 2 mm were correct. 2) regions of 1 mm were wrong, regions of 2 mm were correct. 3) regions of both 1 mm and 2 mm were wrong.

Figure 5

1–6 are respectively: radiomics-tumoral; radiomics-peritumoral 2 mm; radiomics-combined 2 mm; deep learning-tumoral; deep learning-peritumoral 2 mm; deep learning-combined 2 mm models. (A) ROC curve and AUC of the radiomics models, including 1–3. (B) ROC curve and AUC of the deep learning models, including 4-6. (C) Differences among all models in accuracy, sensitivity, and specificity. *Statistical significance.

Figure 6

Forest map displayed the AUC of the three models in subgroup analysis.