EUVIMEDEUVIMED entdecken
Crossref

Reliable Benchmarking of Breast Ultrasound Lesion Classification Requires Patient-Level Validation

Lulu Wang

Journal of Engineering and Science in Medical Diagnostics and Therapy · 01.11.2026 · 9 (4)

DOI: 10.1115/1.4071712 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Engineering and Science in Medical Diagnostics and Therapy, online erschienen am 11.05.2026. Verfasst von Lulu Wang. Die Arbeit nennt 29 Literaturangaben.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Breast ultrasound is widely used for lesion characterization, yet reported deep-learning performance varies substantially with dataset composition, preprocessing, and validation design. A major source of bias arises when patient-level separation is not enforced, allowing correlated images from the same subject to inflate performance estimates. This study presents a leakage-aware benchmark of convolutional neural networks (CNNs), a Vision Transformer (ViT), and a CNN–transformer late-fusion configuration for benign-versus-malignant breast ultrasound classification on the BUS-UCLM dataset. After exclusion of normal-category images, the final cohort comprised 264 images from 36 patients, including 174 benign and 90 malignant images. Seven CNN-family models, one ViT baseline, and one ResNet18–ViT probability-level late-fusion configuration were evaluated using strict patient-level fivefold cross-validation. Additional analyses included fusion ablation, paired Wilcoxon signed-rank testing, gradient-weighted class activation mapping (Grad-CAM) visualization, and an image-level leakage demonstration. Under strict patient-level evaluation, performance was moderate across all models. GoogLeNet achieved the highest mean accuracy (61.48%), InceptionV3 achieved the highest mean macro-F1 (59.24%), and ResNet50 achieved the highest mean area under the receiver operating characteristic curve (AUC) (0.6699), whereas the standalone ViT showed weaker overall discrimination. The late-fusion configuration remained competitive in threshold-dependent metrics but did not surpass the strongest CNN baselines in AUC. Overall, no architecture demonstrated a clear advantage across both threshold-dependent and threshold-independent metrics. By contrast, image-level splitting substantially inflated apparent performance, underscoring the importance of rigorous patient-level separation for credible benchmarking in breast ultrasound classification.

Zitieren

APA

Wang, L. (2026). Reliable Benchmarking of Breast Ultrasound Lesion Classification Requires Patient-Level Validation. Journal of Engineering and Science in Medical Diagnostics and Therapy, 9(4). https://doi.org/10.1115/1.4071712

Vancouver

Wang L. Reliable Benchmarking of Breast Ultrasound Lesion Classification Requires Patient-Level Validation. Journal of Engineering and Science in Medical Diagnostics and Therapy. 2026;9(4). doi:10.1115/1.4071712

BibTeX
@article{Wang2026,
  author = {Wang, Lulu},
  title = {Reliable Benchmarking of Breast Ultrasound Lesion Classification Requires Patient-Level Validation},
  journal = {Journal of Engineering and Science in Medical Diagnostics and Therapy},
  year = {2026},
  volume = {9},
  number = {4},
  doi = {10.1115/1.4071712}
}
Kostenlos starten →