In this talk, I will present results showing the performance of MSE, PCA->LDA, PCA->XGBoost, XGBoost and novel Ensemble Scale-Invariant CNN models in the context of tissue classification for cancer diagnosis using Fourier Transform Infrared Spectroscopy biopsy images. I will then discuss the explainability of those models using LIME [1] to reveal wavenumber ranges of particular importance for this classification task, before extending the CRIME framework [2] using custom semi-supervised VAE models to identify clustered explanation patterns. Finally, I will show how these clustered explanation patterns can be used to produce a model-agnostic, fully explainable linear predictive model for classification purposes.
[1] Ribeiro, Marco Tulio, Sameer Singh, and Carlos Guestrin. ‘“Why Should I Trust You?”: Explaining the Predictions of Any Classifier’. arXiv, 9 August 2016. https://doi.org/10.48550/arXiv.1602.04938.
[2] Zaki, Jihan K., Jakub Tomasik, Jade A. McCune, Sabine Bahn, Pietro Liò, and Oren A. Scherman. ‘Explainable Deep Learning Framework for SERS Bio-Quantification’. arXiv, 12 November 2024. https://doi.org/10.48550/arXiv.2411.08082.
