Extraction of Lesion and Tumor Region in Multi-modal Images Using Novel Self-organizing Map-Based Enhanced Fuzzy C-Means Clustering Algorithm

摘要

Analyzing the medical images and segmenting the same for detecting the tumor and lesion regions embedded within the images are quite a tedious process. On performing the task of tumor and lesion region detection, several intricacies arise and two of the major hindrances are time complexity and accuracy level sustainment. Resolving these two issues is the major concern of this paper and the authors have achieved it, which could be verified from the figures of this paper. If the examination of the medical images obtained through modalities such as MRI and CT is clearly processed using an algorithm, preplanning of surgical procedures could be made with ease. The development of such an algorithm is focused by the authors, and the algorithm framed in this research ensemble the working of self-organizing map (SOM) and enhanced fuzzy C-means (EnFCM), and the authors have collectively named the algorithm as SOM-based EnFCM. The proposed algorithm has produced a high peak signal-to-noise ratio (PSNR) value of 60 dB and mean square error (MSE) of 0.06. The time required by the algorithm for processing 71 input slice images acquired through CT and MRI scans is around 6 s, and the overall accuracy exhibited by the algorithm is 48%. This has given a new and a dynamic approach, which could be greatly used by the radiologists in clinical practices. To contest and prove the efficiency of the SOM-EnFCM algorithm, the segmentation results of SOM and EnFCM algorithms while operating individually are compared.