Melanoma Image Processing and Analysis for Decision Support Systems

摘要

Melanoma is the most aggressive form of skin cancer which is responsible for the majority of skin cancer related deaths. Image processing and analysis of melanoma images can result in (better) detection and early diagnosis and therefore reducing the mortality rate. Efficient pre-processing, image enhancement, segmentation, feature extraction and classification techniques have been developed to improve the performance of Computer Aided Diagnosis (CAD) of melanoma images. Border detection of lesions in melanoma images is important in improving the accuracy of CAD systems in detecting melanoma. We have developed a semi-automated algorithm to discriminate the foreground lesion from skin background by clicking on a small subset of the lesion. Implementing the image processing and analysis algorithms for CAD and decision support systems is computationally demanding. However, due to high inherent parallelism of such algorithms, systems with parallel processors could be useful for accelerating but they are energy intensive and costly. Special reconfigurable hardware such as Field-Programmable Gate Arrays (FPGAs) with powerful parallel processing feature can be used for achieving necessary performance of embedded systems with efficient utilization of hardware resources. In order to achieve acceleration of the image processing and analysis algorithms, we implement the most compute-intensive algorithms of the CAD and decision support systems onto FPGA for deploying as an embedded device. A hardware/software co-design approach was proposed for implementing Support Vector Machine (SVM) classifier for classifying melanoma images online. The hybrid Zynq platform was used for implementing the proposed classifier using High Level Synthesis design methodology. The implemented SVM classification system on Zynq demonstrated high performance with low resource utilization and power consumption, meeting several embedded systems constraints. Overall, the hardware implementation on FPGA could be extended in the future for other computationally demanding parts in the process, aiming to reach an efficient real-time decision support system for enhancing early detection of melanoma with high performance and low cost.