Infant Brain MRI Abnormalities Detection Using Deep Learning
DOI:
https://doi.org/10.17762/msea.v70i2.2092Abstract
Magnetic Resonance Imaging (MRI) is considered as a critical tool for the medical investigation of brain. The defect of congenital brain has distinct set of symptoms and impairments which are difficult to identify and classify with the MRI images. Studies has revealed that the rate of women with the infants of abnormal brain in increasing at a high rate. Early identification of the symptoms can help in precise diagnosis of the brain defect and helps to carry out with the effective treatment plan. The literature survey has shown the segmentation of adult brain and not infants of month’s year old. In this proposed system, steps of processes are proposed for infant brain classification which uses deep learning technique. The significant contribution of this proposed system is to diagnosis the defective brain at the early stage on the infant’s brain development. The proposed system has four phase of pre- processing (filtering noise), enhancement, Feature extraction, CNN based segmentation and classification using the trained network. The constructed algorithm does the gray level conversion of the test image selected. In pre- processing stage, removal of noise takes place and it is followed by the image enhancement using Histogram equalization filter and IM filtering. Gray Level Co-Occurrence Matrix (GLCM) function extracts the feature from the filtered image output. Convolution Neural Network (CNN) does the classification, detection and the segmentation of the image using the trained datasets. The Deep learning based classification and segmentation can improve the prediction accuracy and reduce generalization errors. The all the test image results is updated in a web page with the time stamp using an IoT module for the accurate patience’s survey reporting and other further future analysis. Our future work aims at transfer learning method in which the algorithm concentrates on automatically solving different problems from the knowledge gained while solving the previous set of problems and also improving the output efficiency using more disenable data sets.