Early detection of MS in fMRI images using deep learning techniques

Introduction & Objective:MS is a disease of the central nervous system in which the body makes a defensive attack on its tissues. The disease can affect the brain and spinal cord, causing a wide range of potential symptoms, including balance, movement and vision problems. MRI and fMRI images are a very important tool in the diagnosis and treatment of MS. The aim of this study was to provide an intelligent method based on fMRi image processing that has the ability to detect MS from these images early.   Materials & Methods:This is an experimental study and B RATS data set has been used to diagnose MS, which has 145 folders for patients in different MRI imaging conditions. This data set consists of 4 versions from 2012 to 2018, which in each version of the data set is more and their quality is higher. The main data of this data set is in DICOM format, which in this research has been converted to JPEG format for easier use through DICOM VIEWER software. This research uses 1000 image input samples to confirm the proposed approach and evaluate it.     In this study, a four-layer deep learning network based on convolutional neural network in the 2015b matlab environment is simulated in a system with i7 processor specifications with 6 MB and 3.6 MHz cache and 6 GB of RAM in Windows 10.In the deep learning structure, which itself had the ability to extract features, we used another method to do this. Because deep learning, although capable of extracting features, did so by chance. In order for the previous steps to be definitive, we used another algorithm inside the repeating loops and inside the torsion layer (CONVLOTION) to reduce the dimensions of the feature during the training, secondly to select the best features and thirdly to extract the features. Definitely do. The image was first inserted into the input layer, then into the hidden layer, which includes a twist layer, two POOLING layers, and a connected layer. In the first part of the torsion layer, initial training was given and then it entered the pooling layer. This layer consists of two parts, the first is the max pooling layer and the second is the random pooling layer. The segmentation operation was performed in the max pooling layer and the feature extraction was performed in the random pooling layer. It should be noted that in both pooling layers, we used a 3 * 3 window filter to perform the segmentation and feature extraction operations well. Then, the output of the pooling layer is entered into the continuous layer after fragmentation and extraction of features. In this fully connected layer, the data were tested with the help of 2 * 3 window filter and at the end, the final classification for different modes was displayed. Results: The graphical representation of the ROC curve showed that the degree of sensitivity or correct prediction versus incorrect prediction in this binary classification system in which the separation threshold varies was significant. The area below this curve AUC = 8592/0 was the same as the accuracy of the proposed method was 98.689% and the sensitivity was 94.8766%.   Conclusion:Given the prevalence of MS, early diagnosis and presentation of an intelligent method based on fMRI imaging is essential for treatment. This intelligent method tries to be able to help in diagnosis and treatment with more accuracy, better identification of features and patterns affecting the disease than previous methods as physicianschr('39') assistants. Finally, the results of this study showed that the efficiency of the proposed method was evaluated at a high level And showed its optimality as much as possible. In addition, the results showed the speed of training and testing of data in high volume and fast convergence of the algorithm. It is also easier to expand and generalize.