A Reinforcement Learning Algorithm for Automated Detection of Skin Lesions

Skin cancers are increasing at an alarming rate, and detection in the early stages is essential for advanced treatment. The current segmentation methods have limited labeling ability to ground truth images due numerous noisy expert annotations present datasets. precise boundary correctly locate diagnose various skin lesions. In this work, lesion method proposed as a Markov decision process. It solved by training agent segment region using deep reinforcement-learning algorithm. Our similar delineation of interest physicians. follows set serial actions delineation, action space defined continuous parameters. model learns deterministic policy gradient enables improvement performance we proceed from coarse results finer results. Finally, our evaluated on International Imaging Collaboration (ISIC) 2017 image dataset, Human against Machine (HAM10000), PH2 dataset. On ISIC algorithm achieves accuracy 96.33% naevus cases, 95.39% melanoma 94.27% seborrheic keratosis cases. other metrics these datasets rank higher when compared with state-of-the-art algorithms.