Multinomial classification of CT-MRI image retrieval by optimizing EffResNet-4 architecture in deep neural models

Medical imagining in healthcare is a state-of-art approach for the identification of injuries in cells of the brain or the detection of abnormal structures treatment of patients. The major steps involve precise identification, segmentation, and detection of damaged cells in the brain MRI imaging in a timely manner. Segmentation in the case of multiple classes faces challenges such as scaling, training, and testing accuracy, time consumption, etc. This paper implements the fully convolutional neural network that utilizes the EfficientNet model with its variant B4 from the EfficientNet family. The purpose of the model selection is grounded on the capability via scaling the input image in three dimensions in order to achieve accuracy with ensembling the model with ResNet architecture for semantic segmentation for the pre-trained network. The substantial network will provide robustness in the feature engineering process via encoder and decoder from EfficientNet and ResNet neural models respectively. The ensemble model named EffResNet-4 will perform neural augmentation to provide a meta-learning approach for training and validation in analysing the abnormal structures in the multi-class classification of medical images. The proposed model confront the trade-off between performance and model size, providing a more adaptable and potent solution for segmentation and detection for medical imaging. The experimented findings reveal that the proposed methodology has 3064 images with three classes in it which quantitatively score 0.81 mean dice score and 99.39% accuracy in training and it is compared to other competitive neural models.