Multiple instance learner for prognostic tissue pattern identification

Abstract

The invention relates to a method (100) of identifying tissue patterns being indicative of a patient-related attribute value. The method comprises: receiving (102) digital images (212) of tissue samples of patients, the images having assigned a label indicating a patient-related attribute value; splitting (104) each received image into a set of image tiles (216); computing (106) a feature vector (220) for each tile; training (108) a Multiple-lnstance-Learning (MIL) program (226) on all the tiles and respective feature vectors for computing for each of the tiles a numerical value (228) being indicative of the predictive power of the feature vector associated with the tile in respect to the label of the tile's respective image; and outputting a report gallery (206) comprising tiles sorted in accordance with their respectively computed numerical value and / or comprising a graphical representation of the numerical value.

Description

technical field [0001] The present invention relates to the field of digital pathology, more particularly to the field of image analysis. Background technique [0002] Several image analysis methods are known which can be used to aid in diagnostic procedures and to identify appropriate treatments based on the analysis of images of tissue samples. [0003] Some image analysis techniques are based on using different programs to search for structures in images that are known to serve as indicators of the presence of a particular disease and / or the likelihood of successfully treating that disease with a particular drug. For example, some drugs used during immunotherapy in cancer patients only work if certain immune cells are present at a certain distance from the cancer cells. In this case, an attempt is made to automatically recognize these objects in tissue images, ie certain cell types or certain subcellular and supracellular structures, in order to be able to make a stateme...

AI Technical Summary

Problems solved by technology

The downside of these methods is that they usually work like a black box

In other words, pathologists using these techniques must rely on the predictive power of these algorithms without being able to pinpoint which tissue trait is ultimately decisive for the prediction

This can be a significant disadvantage, e.g. in drug approvals, since for this purpose the patient population that benefits from a certain treatment must be clearly indicated

Images