Examination of red blood cell morphology is one of the diagnostic aids for several diseases, one of which is anemia. The development of the application of digital image processing technology, artificial intelligence, and computer-assisted diagnosis opens opportunities to solve various problems related to medical images. Red blood cells sticking together or overlapping is a challenge in the red blood cell segmentation process which ultimately affects the results of cell type identification. A method that can perform instance segmentation is needed to overcome this problem. This study aims to implement the Mask-RCNN algorithm on a small red blood cell image dataset and evaluate the prediction results' performance. Based on the research results, the attached red blood cells can be detected individually by the model, and the accuracy of the cell detection results is 68.27%. Mask-RCNN can be used for blood cell segmentation instances and blood cell detection on small datasets, but the model accuracy still needs to be improved. Therefore it is necessary to do further research by increasing the number of datasets used.
Copyrights © 2023