Anomaly detection for medical samples under multiple settings

Abstract

Using multiple imaging modes in whole slide image screening is potentially useful to reduce false positives. To use multiple imaging modes, a method for locating anomalies on a medical sample from an image thereof uses an anomaly-detection process that comprises using plural base classifiers individually to classify an object-of-interest suspected to be an anomaly. Each base classifier respectively extracts features of the object-of-interest and generates, according to the extracted features, a score indicating a likelihood of the object-of-interest being anomalous. The anomaly-detection process further comprises using an aggregate classifier to combine the scores generated by the base classifiers to determine whether the object-of-interest is the anomaly. The aggregate classifier determines a dependability measure for each base classifier according to setting-based variables of a setting under which the sample and the image are obtained, and then selectively combines the scores of the base classifiers according to the dependability measures.