The chapter presents a method for diagnosing oncourological diseases based on machine learning algorithms. MobileNet50, ResNet50 convolutional neural networks are used to solve the problem of classifying patient biopsy image segments according to the Gleason scale. Augmentation technologies were applied to the existing data set for better performance of the neural network. The accuracy of the algorithm was estimated by the total error and the Cohen's Kappa coefficient. The results of the algorithm in software show a good level of accuracy: in 65% of cases, the algorithm accurately determined the Gleason index, and the rest of the data had a slight deviation of the confusion matrix.
TopIntroduction
Oncourological diseases are one of the most significant medical and social problems both in Russia and in all other countries of the world. They are one of the major causes of mortality and disability of the population (Siegel et al., 2021), which leads to a significant loss of the able-bodied part of society. Among urological oncological diseases, according to the number of diagnosed and fatal cases, prostate cancer can be distinguished. According to statistics, prostate cancer is in second place in the number of cancer deaths among men (World Health Organization, 2016).
Early diagnosis of oncourological diseases is possible, since structural changes in the basic biochemical substances of cells begin much earlier than the clinical manifestation of the symptoms of a malignant tumor. This is facilitated by the technology of modern recognition of cancer cells. Not only the detection of cancer but also the determination of the stage of the disease have an important role in the treatment. With an incorrectly defined stage of the disease, premature surgical operations are possible, which usually limits the further full-fledged life of a person. When detecting prostate cancer, for example, based on the level of cancer markers, it is necessary to correctly determine the current stage of the oncological disease. For this purpose, the tissue of the affected organ, the prostate, is examined directly.
In medical practice, various methods of diagnosing prostate cancer are used, while the only accurate method of detecting and determining the degree of cancer is prostate biopsy analysis (Frankel et al., 2003). A prostate biopsy is performed for histological diagnosis of cancer and for definitive diagnosis. It also allows you to determine the degree of aggressiveness of the tumor and the stage of the disease (its prevalence). The results of a prostate biopsy are the most important factor that determines the tactics of treating the patient, as well as the prognosis of the disease.
The Gleason scale is used to identify prostate cancer (Epstein, 2010), according to which prostate carcinomas are classified.
Different histological samples are assigned numbers from 1 (well differentiated) to 5 (poorly differentiated).
The final score on the Gleason scale is the sum of the two dominant patterns, and in modern clinical practice the lowest score is 6 points (3 + 3).
Histological evaluation of human tissue based on visual analysis of tissue sections takes a long time and is often of limited quality. In prostate cancer in particular, it is difficult to unequivocally determine the difference between the Gleason scale 3 and 4 scores. Therefore, mistakes are often made when determining the current degree of prostate cancer.
Thus, the development of new methods and approaches of digital medicine that increase the speed and accuracy of diagnosis is an important task.