Development and application of artificial intelligence for the detection of the lymph nodes metastases in colorectal cancer

Abstract

AIM: to create a marked data set (histoscans of lymph nodes) for use in the development of medical decision support systems (based on machine learning) in pathomorphology, which will allow determining the presence of metastatic lymph node lesions in CRC. RESULTS: the dataset included 432 files with digital images and markings of 1019 lymph nodes, including lymph nodes with and without metastases. Based on the marked-up data, a neural network model was trained to determine the probability of metastatic lesion for each pixel in the area of interest — the lymph node (Dice 0.863 for the replaced tissue, Dice macro 0.923). In addition, pre-and postprocessing methods were implemented to represent input data in a form acceptable for machine learning and to represent the AI model’s response in a form convenient for user perception. Additionally, a neural network model has been developed that predicts the probability of finding artifacts in digital images of lymph nodes with the possibility of forming an artifact probability map (Dice macro0.776; Dice for artifacts 0.552; IoU macro 0.725 and IoU for artifacts 0.451). CONCLUSION: the developed model is a good basis for the implementation of a full-fledged solution, on the basis of which a system can be developed to assist doctors in finding and evaluating the replacement of tissue structures and determining metastatic lymph node lesions, detecting artifacts and evaluating the quality of digital images.