Deep Learning for Automated Detection of Cysts and Tumors: A Review

Medical imaging involves the detection of cysts in order to enable early diagnosis and treatment of a disease. The conventional methods of diagnosis are based on subjective interpretation of radiologists that is time consuming. The latest development of machine learning and deep learning make possible the automated and highly accurate cyst detection. The review entails the discussion of both ML and DL techniques of cyst detection and segmentation in ultrasound, MRI, CT and optical coherence tomography. It cross-cuts through the popular datasets, preprocessing mechanisms, feature extraction mechanisms and classifiers including Convolutional Neural Networks, Random Forest and Support Vector Machines. It further evaluates the state of art approaches, such as transfer learning, deep supervision, and integration of domain knowledge in improving the segmentation performance. The significant concerns that cover the lack of adequate data, explainability of the model, and clinical adoption are discussed. The next research directions include Explainable Artificial Intelligence, federated learning and multimodal methods of improving the accuracy and reliability of cyst detection in real-world clinical environment.