Proposing a hybrid method for mitotic figure detection, counting, and Ki67 index calculation in the grading of histopathological images of neuroendocrine tumors

Neuroendocrine tumors (NETs) result from the abnormal growth and uncontrolled proliferation of neuroendocrine cells, and pathologists examine them under a microscope to prepare reports. These reports determine the type, grade, and rate of spread of NETs, shaping the patient's treatment plan. The rapid and accurate evaluation of pathological images is crucial for the detection and treatment planning of the disease. The use of artificial intelligence-based methods for the assessment of pathological processes in NETs can reduce the workload of pathologists. Additionally, in regions with a shortage of expert pathologists, artificial intelligence can play a guiding role. As part of this thesis, with the approval of the ethics committee report and the assistance of expert pathologists, images were obtained from preparations using two different staining methods to diagnose NETs. Firstly, the YOLOv5 architecture was used for mitotic figure detection from H\&E stained images, and a YOLOv5-transformer model was implemented by adding a transformer structure to this architecture. Secondly, a hybrid approach combining U-Net deep learning algorithms and traditional image processing methods was proposed for calculating the Ki67 proliferation index from IHC stained images. The use of the proposed methods will reduce the workload of pathologists, expedite processes, and minimize the risk of overlooking details. Moreover, the ability of these methods to perform well independently of the amount of data will allow their use in clinical settings. In the literature, no artificial intelligence-supported study combining mitotic figure detection, counting, and Ki67 proliferation index assessment for NET evaluation has been encountered. In this regard, it serves as a guiding study in the field of pathology for NET evaluation. Simultaneously, it provides a resource that can be used for educational purposes for less experienced pathologists and students.