Vaidya, Nikhil (2022) A Comparative Study To Classify Malignant Skin Lesion Using Mask-RCNN. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (1MB) | Preview |
Preview |
PDF (Configuration manual)
Download (1MB) | Preview |
Abstract
Rapidly progressing melanomas of the skin can be lethal if they go undetected for an extended period. Recent studies have found that 1/3 of all cancers are skin cancers. If it remains undiagnosed, the disease can cause severe harm or even be fatal for the patient. Thus, it is essential to detect melanoma early and begin treatment immediately. Conventional medical diagnosis typically requires painful and expensive skin samples taken from the patient. Melanoma may be detected quickly and effectively using contemporary AI (Artificial Intelligence) and deep learning techniques. This study employs a novel method for melanoma identification by applying automated custom annotation on Mask R-CNN with ResNet101/50 backbones to classify lesions. The publically accessible ISIC Dataset (International Skin Imaging Collaboration) is used to perform this task. Mean accuracy and recall (mAP and mAR) ratings are used to grade the model. Using our model, we got an mAP score of 78.6 and an mAR of 97.6. The effectiveness of the models is evaluated via a thorough comparison.
Actions (login required)
View Item |