Machine learning with different digital images classification in laparoscopic surgery

Abstract

The evaluation of the effectiveness of the automatic computer diagnostic (ACD) systems developed based on two classifiers – HAAR features cascade and AdaBoost for the laparoscopic diagnostics of appendicitis and ovarian cysts in women with chronic pelvic pain is presented. The training of HAAR features cascade, and AdaBoost classifiers were performed with images/ frames, which have been extracted from video gained in laparoscopic diagnostics. Both gamma-corrected RGB and RGB converted into HSV frames were used for training. Descriptors were extracted from images with the method of Local Binary Pattern (LBP), which includes both data on color characteristics («modified color LBP» - MCLBP) and textural characteristics, which have been used later on for AdaBoost classifier training. Classification of test video images revealed that the highest recall for appendicitis diagnostics was achieved after training of AdaBoost with MCLBP descriptors extracted from RGB images – 0.708, and in the case of ovarian cysts diagnostics – for MCLBP gained from RGB images – 0.886. Developed AdaBoost-based ACD system achieved a 73.6% correct classification rate (accuracy) for appendicitis and 85.4% for ovarian cysts. The accuracy of the HAAR features classifier was highest in the case of ovarian cysts identification and achieved 0.653 (RGB) – 0.708 (HSV) values. It was concluded that the HAAR feature-based cascade classifier turned to be less effective when compared with the AdaBoost classifier trained with MCLBP descriptors. Ovarian cysts were better diagnosed when compared with appendicitis with the developed ACD.