Efficient Out-of-Distribution Detection of Melanoma with Wavelet-Based Normalizing Flows

Melanoma is a serious form of skin cancer with high mortality rate at later stages. Fortunately, when detected early, the prognosis melanoma promising and malignant incidence rates are relatively low. As result, datasets heavily imbalanced which complicates training current state-of-the-art supervised classification AI models. We propose to use generative models learn benign data distribution detect Out-of-Distribution (OOD) images through density estimation. Normalizing Flows (NFs) ideal candidates for OOD detection due their ability compute exact likelihoods. Nevertheless, inductive biases towards apparent graphical features rather than semantic context hamper accurate detection. In this work, we aim using these domain-level knowledge melanoma, improve likelihood-based images. Our encouraging results demonstrate potential NFs. achieve 9% increase in Area Under Curve Receiver Operating Characteristics by wavelet-based This model requires significantly less parameters inference making it more applicable on edge devices. The proposed methodology can aid medical experts diagnosis skin-cancer patients continuously survival rates. Furthermore, research paves way other areas oncology similar imbalance issues (Code available at: https://github.com/A-Vzer/WaveletFlowPytorch ).