Liquid Neural Networks (LNN) for Early Breast Cancer Diagnosis — 26a — Chenchaiah Mekalathuru, Dr. KC Santhosh, Dr. Rodrigue Rizk

Breast cancer remains the leading cause of cancer-related deaths among women globally, with survival rates varying dramatically across different economic regions. In response, the World Health Organization launched the Global Breast Cancer Initiative (GBCI), focusing on enhancing health promotion, early detection, timely diagnosis, and comprehensive management of breast cancer.  Addressing the critical need for early breast cancer detection, this paper proposes the application of Liquid Neural Networks (LNNs), which leverage Neural Circuit Policies (NCPs) to optimize model  efficiency. Unlike traditional architectures such as Deep Neural Networks (DNNs) and ResNets, which, despite their effectiveness, require extensive computational resources due to their large parameter counts, LNNs offer a substantial reduction in parameter complexity while maintaining high accuracy, making them ideally suited for clinical use. Our experimental findings on the breastMNIST dataset reveal that LNNs exceed the performance of established models like ResNets and classical  DNNs, achieving an Area Under the Curve (AUC) score of 0.901 and an accuracy rate of 0.884. These results underscore the potential of LNNs to transform breast cancer diagnostics by offering high accuracy classification with greater computational efficiency, suggesting that their integration into clinical workflows could significantly improve diagnostic processes in diverse healthcare  environments.

University of South Dakota
Dr. Rodrigue Rizk