فصلنامه تخصصی فناوری اطلاعات و ارتباطات
نوع مقاله : مقاله پژوهشی
نویسندگان
1 گروه مهندسی کامپیوتر، واحد علوم و تحقیقات، دانشگاه آزاد اسلامی، تهران، ایران
2 گروه مهندسی کامپیوتر، واحد فیروزآباد، دانشگاه آزاد اسلامی، فیروزآباد، ایران
3 گروه مهندسی برق، واحد علوم و تحقیقات، دانشگاه آزاد اسلامی، تهران، ایران
4 دانشکده مهندسی برق و کامپیوتر، واحد اصفهان (خوراسگان)، دانشگاه آزاد اسلامی، اصفهان، ایران
کلیدواژهها
عنوان مقاله English
نویسندگان English
Background and Objectives: Multiple sclerosis is a brain disease where early diagnosis is crucial for treatment. One of the ways to diagnose this disease is by observing lesions caused by it in MRI scans. Most previous approaches have issues such as low diagnostic accuracy, a high number of features, time-consuming analysis, and a lack of certainty in achieving optimal answers.
Methods: In this article, for the first time, a feature vector set is formed by aggregating results from image MRI texture descriptors such as wavelet transform, chaotic features (fractal), and local binary patterns. The presentation of a selected feature set using a differential evolutionary algorithm has not been utilized in this area of identification before, so our proposed technique is based on this approach. Additionally, the proposed classifier will be an improved model combining three types of neural networks. Moreover, improvements in accuracy, sensitivity, and the ability to verify the correctness of the classification results are also considered innovative aspects.
Findings: The data used in this article was obtained from two datasets. After K-fold cross-validation, the experimental accuracy for both image datasets were found to be 95% and 97%, respectively, which represents a 2% improvement over a method that used wavelet transform along with principal component analysis and support vector machines, while also addressing the uncertainty issue.
Conclusion: Our integrated algorithm introduced greater diagnostic accuracy compared to previous methods and takes into account the accuracy factor that was not considered in past approaches. This algorithm not only reduces processing time but also enables simultaneous processing from different channels, aligning better with the opinions of specialized doctors. Despite the lack of a simultaneous separation-processing technique, the rates of false positives and negatives in identifying MS disease are very low. For future work, recommendations include noise removal in the preprocessing stage, combining feature selection techniques to increase accuracy, and using parallel processing as the primary tool in separation software.
کلیدواژهها English
