Shape Analysis of AF Segments for Rapid Assessment of Mohs Layers for BCC Presence by AF-Raman Microscopy

An automated method to detect Basal Cell Carcinoma (BCC) relies on Autofluorescence (AF) imaging guiding Raman microscopy to obtain biochemical information for tissue classification. The guidance is provided via an image segmentation technique aiming to reduce the risk of missing cancer. We present evidence that shape of an AF segment may be useful for ‘trimming’ 6–8% of non-BCC segments in an essentially coordinate free manner, without compromising BCC detection. By allowing the AF-Raman method to direct the more time consuming Raman analysis toward more relevant regions, the proposed trimming of unnecessary segments should ultimately improve the overall accuracy. The presented shape analysis uses the recently introduced Weighted Euler Curve Transform (WECT). WECT embeds segments in a space of real matrices of fixed dimensions, where a shape is an equivalence class of segments with WECTs matching under a cyclic permutation of columns. The induced rotation invariant distance is non-Hilbertian, which requires special care in using it with kernel methods (e.g. Kernel PCA, Kernel LDA, SVMs). Our currently best results are achieved by \(L_1\) SVMs based on the Laplace ‘kernel’.