Frontiers in oncology · 2026
Deep transfer learning for breast cancer detection in underserved regions.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Obaid Mahmoud, Odeh Suhail M, Ashqar Huthaifa I, Abumwais Allam, Hodrob Rami
Аннотация
BACKGROUND AND OBJECTIVE: In Palestine, breast cancer is the leading cancer in women, constituting more than 34% of all cancer cases in women and 12% of all cancer related deaths. The Palestinian health care system is suffering from shortages, lack of advanced diagnostic equipment, among other issues. This study proposes a new two-step deep learning method for breast cancer detection in mammograms, with a particular focus on the potential of using it in low-resource settings like Palestine. The framework categorizes the tumor and adds a cost-effective and scalable diagnostic support tool that differentiates benign from malignant tumors.
METHODS: The proposed framework operates in two sequential stages. First, a U-Net architecture with a VGG16 encoder backbone, trained from scratch on the CBIS-DDSM mammography dataset, performs lesion segmentation. The CBIS-DDSM training set comprised 2,206 mammograms and the test set 576 images, all with ground truth binary masks. Second, a VGG16 classification model initialized with ImageNet pretrained weights classifies the segmented regions as benign or malignant. The classifier was tested on a subset of 34 patients from the Palestine Hospital dataset (12 benign, 22 malignant), which was used exclusively for external testing, in which no Palestine data were used in training or validation.
RESULTS: The U-Net segmentation model achieved a mean IoU of 0.70, Dice coefficient of 0.74, precision of 0.78, and recall of 0.71 on the CBIS-DDSM test set. The VGG16 classifier achieved 91% accuracy, 0.91 precision, 0.95 recall (malignant class), and AUC of 0.97 on the Palestine evaluation subset. Comparison against ResNet50 (85% accuracy) and MobileNet (82%) confirms the superiority of the proposed approach.
CONCLUSION: The present study proposes a promising proof-of-concept deep learning pipeline for breast cancer detection. The results show that it is feasible in a low-resource environment, but would need to be validated on a larger annotated local dataset before deployment. The framework provides a guidance for scalable and cost-effective diagnostic assistance in underserved areas.
METHODS: The proposed framework operates in two sequential stages. First, a U-Net architecture with a VGG16 encoder backbone, trained from scratch on the CBIS-DDSM mammography dataset, performs lesion segmentation. The CBIS-DDSM training set comprised 2,206 mammograms and the test set 576 images, all with ground truth binary masks. Second, a VGG16 classification model initialized with ImageNet pretrained weights classifies the segmented regions as benign or malignant. The classifier was tested on a subset of 34 patients from the Palestine Hospital dataset (12 benign, 22 malignant), which was used exclusively for external testing, in which no Palestine data were used in training or validation.
RESULTS: The U-Net segmentation model achieved a mean IoU of 0.70, Dice coefficient of 0.74, precision of 0.78, and recall of 0.71 on the CBIS-DDSM test set. The VGG16 classifier achieved 91% accuracy, 0.91 precision, 0.95 recall (malignant class), and AUC of 0.97 on the Palestine evaluation subset. Comparison against ResNet50 (85% accuracy) and MobileNet (82%) confirms the superiority of the proposed approach.
CONCLUSION: The present study proposes a promising proof-of-concept deep learning pipeline for breast cancer detection. The results show that it is feasible in a low-resource environment, but would need to be validated on a larger annotated local dataset before deployment. The framework provides a guidance for scalable and cost-effective diagnostic assistance in underserved areas.