| Malignant melanoma is the most lethal form of skin cancer and is responsible for 75% of all deaths from skin cancer. However, malignant melanoma can be treated successfully if detected in the earliest stage.; This project deals with the design and implementation of a hybrid artificial intelligence system, which can help to diagnose malignant melanoma. In this study, a three-stage model is used in the classification system that combines image-processing techniques, expert systems, neural network systems, and fuzzy inference systems to diagnose skin cancer. In Stage I, image-processing techniques are used to extract features that characterize skin tumors; a feature selection process based on the correlation analysis and aided with decision tree is also designed and implemented. In Stage II, the neural networks work as classifiers that are used to classify input images into different tumor categories. Two different hybrid neural network architectures are developed—a voting scheme and a hierarchical structure. In Stage III, the outputs from neural networks are combined by a fuzzy inference system to obtain an improved performance. In this system, an optional expert system which consists of a set of rules based on the statistics of skin tumors and the clinical experience of a dermatologist is used. Furthermore, these two hybrid architectures are combined to achieve the best performance.; This computer-aided diagnostic system is trained and tested with a large number of lesions and promising results are demonstrated. This system also provides a general framework for a hybrid classification system. |
