Reducing Global Breast Cancer Mortality with Early, Low-Cost Diagnosis Using Generative AI and Hybrid DL Architecture

Breast cancer is the most common cancer in the world [1]. Mortality rates stand at 27 percent in India and 60 in Africa, compared to 7 in the US [2]. This disconnect boils down to the time of diagnosis. According to the CDC late-stage diagnosis has a five-year mortality rate of 69 percent, compared to 2 percent for early diagnosis [3]. Underserved communities lack access to high-cost, advanced diagnostic tech-nologies (MRIs/Mammograms) and cancer specialists (Oncolo-gists/Radiologists) [4]. Patients must depend on low-cost, low-resolution modalities (Ultrasounds) and the limited breast cancer expertise of general physicians (and gynecologists in developing countries), significantly delaying diagnosis and subsequent in-tervention [5]. This study aims to reduce global breast cancer mortality and healthcare costs by enabling accurate, early, and low-cost diagnosis using the existing ultrasound infrastructure in developing countries and underserved communities. To address data scarcity and skew, advanced image augmentation techniques such as deep convolutional GANs and stable diffusion models were evaluated. In each case of data augmentation, two classification approaches were compared: a lightweight MobileNetV2 convolutional neural network and a hybrid model combining ResNet50 for feature extraction with statistical machine learning models. The highest detection accuracy (95 percent) was achieved using the hybrid model with ResNet50 and XGBoost, paired with Stable Diffusion augmentation. An iOS app was implemented to demonstrate the feasibility of this approach in resource-constrained settings, showcasing its potential for scalable, efficient diagnostic solutions.