BREAST AI: Low Cost, Explainable Artificial Intelligence Based App for Efficient Diagnosis of Breast Cancer in Developing Areas

Abstract

Breast cancer is the most common cancer among women. Accurate diagnosis of breast cancer is crucial for successful treatment. However, many women in developing areas do not have access to similar medical resources as those in more developed areas, resulting in poor outcomes. Disparities range from a lack of mammograms in rural settings, as they are solely applicable to the breast, to a lack of trained radiologists to diagnose breast cancer. The goal of this research was to develop an artificial intelligence-based app for the accurate diagnosis of breast cancer via ultrasound scans without the need for an internet connection. The artificial intelligence model developed was a fine-tuned MobileNet-V2 convolutional neural network, trained on 624 ultrasound scans and tested on 156 scans from the Baheya Hospital in Egypt. The testing accuracy was 87.50%. To understand how the algorithm was making predictions (Explainable AI), code was created to generate heat maps to highlight parts of the breast ultrasound that were most significant in helping the algorithm make its decision. The AI model was downloaded as a TensorFlow Lite file, which allows for on-device machine learning, eliminating the need for an internet connection to operate, and developed into a user-friendly app. Further work, such as using a more diverse dataset, trying different convolutional neural network architectures, building a multimodal model, and implementing denoising algorithms can be done to improve this research.

Ähnliche Arbeiten