Yapay zekâ ile mesane kanserinin lenf nodu metastaslarının otomatik tanınması

Deep learning, which is an important subheading of artificial intelligence, enables significant improvements and facilitations in healthcare services. In this scope, health-related applications developed by deep learning methods are increasing day by day. Studies such as the detection of cancer cells, medical imaging, predicting diseases and recording diseases in cloud systems can be given as examples. Detection of diseased and cancerous cells is of great importance. Since "early diagnosis saves lives" is the most important medicine rule, early diagnosis of cancer cells through artificial intelligence makes a significant contribution to getting patients back to life. There are several different types of cancer, and bladder cancer is one of the most common types of cancer. Bladder cancer, which involves small tumor cells as well as cancer cells that spread over a wider area of tissue, is a difficult form of cancer to diagnose. The aim of this study is to detect the lymph node metastasis of bladder cancer using artificial neural networks. In experiments, while cancer cells that spread across large areas can be identified to a large extent by a deep learning model, this performance is lower in tumors that spread to small areas. The main purpose of this study is to achieve successful diagnosis accuracy in small cancer areas as well as in large cancer areas. Thus, significant convenience will be achieved with the recognition of small tumor areas in which doctors who have long, challenging working hours and quite high number of patients can overlook. Specially created dataset for the study was tested with the Mask R-CNN algorithm using the ResNet model. In the test results, it was observed that while significant success was achieved in tumor cells which covers large areas, the model did not achieve the same success on the small areas. For this purpose, changes and additions were made in the number of layers and stages of the model and a significant improvement in small tumor area detection was obtained.