In the medical field, it is delicate to anticipate good performance in using deep learning due to the lack of large-scale training data and class imbalance. In particular, ultrasound, which is a key breast cancer diagnosis method, is delicate to diagnose accurately as the quality and interpretation of images can vary depending on the operator’s experience and proficiency. Therefore, computer-aided diagnosis technology can facilitate diagnosis by visualizing abnormal information such as tumors and masses in ultrasound images. In this study, we implemented deep learning-based anomaly detection methods for breast ultrasound images and validated their effectiveness in detecting abnormal regions. Herein, we specifically compared the sliced-Wasserstein autoencoder with two representative unsupervised learning models autoencoder and variational autoencoder. The anomalous region detection performance is estimated with the normal region labels. Our experimental results showed that the sliced-Wasserstein autoencoder model outperformed the anomaly detection performance of others. However, anomaly detection using the reconstruction-based approach may not be effective because of the occurrence of numerous false-positive values. In the following studies, reducing these false positives becomes an important challenge.

The Oncotype DX^® 21-gene recurrence score (RS) guides adjuvant chemotherapy decisions in estrogen receptor-positive, human epidermal growth factor receptor 2-negative (ER+/HER2−) breast cancer, yet requires invasive tissue sampling and involves substantial costs. This study evaluated intratumoral tumor ecological diversity (iTED), a habitat imaging approach, as a non-invasive complement for predicting Oncotype DX^® high-risk classification (RS > 25). This retrospective multi-center study included 312 patients with ER+/HER2− invasive breast cancer who underwent Oncotype DX^® testing (development: n = 168; external validation: n = 144). The iTED framework employed superpixel-based habitat determination using Gaussian mixture models on pretreatment dynamic contrast-enhanced MRI. Four predictive models were compared: clinical, conventional whole-tumor radiomics (C-radiomics), iTED, and combined (Clinical + iTED). The iTED model achieved higher discriminative performance compared with C-radiomics in both development (area under the curve [AUC]: 0.868 ± 0.068 vs. 0.730 ± 0.112) and external validation (AUC: 0.811 vs. 0.587) sets. The combined model further improved performance (development AUC: 0.908 ± 0.043; external AUC: 0.889). Habitat imaging-based iTED features achieved numerically higher performance than conventional radiomics in predicting Oncotype DX^® high-risk classification. These findings suggest the potential of iTED as a non-invasive imaging biomarker to support molecular testing in clinical decision-making.