Methods for the evaluation of quality in machine processing of biomedical images

摘要

The authors report on the main objectives and evaluation methods within microscopic image processing as related to the layer of smooth muscle of the intestinal wall. The presented images were acquired to facilitate effective study of diverticulosis. The general aim of image processing is to determine the number of cell nuclei highlighted by a contrast agent. By enabling comparison of the numbers of cell nuclei established in healthy individuals and in patients suffering from diverticulosis, the entire corpus of processing-related operations will contribute to further development within the recognition and assessment of structural changes in intestinal tissue layers. Image processing can be realized using traditional fast methods, for example thresholding with the subsequent summation of segments. The disadvantages of these methods consist in their inability to recognize overlapping nuclei, nuclei areas disrupted by image artefacts, or tissue microstructures; this deficiency causes a considerable error rate and necessitates manual postprocessing. However, such manual verification of results in a comprehensive image database is time-consuming and produces additional errors brought about by various aspects, for instance the verifying expert's fatigue. The metrics for the evaluation of quality in the obtained processing results consists in the values of correctly identified and incorrectly rejected.