Skripsi
INSTANCE SEGMENTATION PADA CITRA MIKROSKOPIS TELUR CACING PARASIT MENGGUNAKAN METODE CASCADE MASK R-CNN
Helminthiasis is classified by the World Health Organization (WHO) as one of the Neglected Tropical Diseases (NTDs) due to its wide distribution, yet it often remains a low priority in global health agendas. Control efforts for helminthiasis are generally carried out through manual microscopic examination. However, this method requires specialized expertise, has technical limitations such as low sensitivity and reproducibility, and is vulnerable to interpretation errors due to manual reading. This study applies an instance segmentation method using the Cascade Mask R-CNN architecture. The model was trained on 2,200 images consisting of 11 classes of parasitic worm eggs, which were re-annotated from bounding box format into instance segmentation format. The study used 24 model configurations with variations in backbone architecture, learning rate scheduler, and number of iterations. The results show that the best model achieved an mAP50 of 0.990, an mAP75 of 0.980, and an mAP50-95 of 0.931 on segmentation masks. Meanwhile, for bounding boxes, the model achieved an mAP50 of 0.985, an mAP75 of 0.990, and an mAP50-95 of 0.909. These results were obtained using the ConvNeXt-T backbone, a cosine annealing with linear warm-up scheduler, and 13,200 iterations. In addition, this research successfully developed a desktop-based software application for parasitic worm egg segmentation.
No other version available