41747_2019_121_MOESM1_ESM.docx (4.14 MB)
Additional file 1: of Diagnostic performance of machine learning applied to texture analysis-derived features for breast lesion characterisation at automated breast ultrasound: a pilot study
journal contribution
posted on 2019-11-02, 07:48 authored by Magda Marcon, Alexander Ciritsis, Cristina Rossi, Anton Becker, Nicole Berger, Moritz Wurnig, Matthias Wagner, Thomas Frauenfelder, Andreas BossTable S1. Inter-reader agreement for the different TA features was evaluated using the intraclass correlation coefficient (ICC). Figure S1. Correlation matrix generated from the full texture feature set for the sub-datasets lesions versus normal tissue (A) and malignant versus benign solid lesions (B) as well as from the corresponding reduced feature set (C) and (D). A significant co-correlation of several features is present in particular among the higher order features in A (e.g., SRE[GLCM] and HGRE[GLCM]) as possible reflection of underlying common biological properties. Figure S2. Heatmaps depicting the optimal hyperparameters for the full feature (A, B) and the reduced feature training datasets (C, D). The hyperparameter tuning was implemented via nested grid search on the SVM classifier by specifying the parameter for gamma and (C) in a logarithmic scale from 0.00001 to 0.001 and 1 to 1000, respectively. (DOCX 4241 kb)