HomeWhy Challenges?All ChallengesCreate Your Own ChallengeContributorsForum
Sign in / Register


Team: Smart Imaging Technologies Co., US

Authors: Vitali Khvatkov, Alexei Vylegzhanin


We used two-step multiscale analysis with ensemble of statistical learning models trained on structure and phase descriptors from the image patches. First step is selection of candidates at magnification of 4 microns per pixel. At this step, we used texture descriptor (27 features) with ensemble of Support Vector Machine Models to select image patches for further analysis. Second step is analysis of candidates with boosted multiscale model cascade that included classifiers based on handcrafted features and Convolutional Neural Network. To classification results were aggregated with geographic neighborhood clustering methods for submission.


The following figure shows the receiver operating characteristic (ROC) curve of the method.


The following figure shows the free-response receiver operating characteristic (FROC) curve of the method.


The table below presents the average sensitivity of the developed system at 6 predefined false positive rates: 1/4, 1/2, 1, 2, 4, and 8 FPs per whole slide image.


Consortium for Open Medical Image Computing © 2012-