Comparative Study for Image Fusion using Various Deep Learning Algorithms

Abstract

Image fusions are that join medical images from many modalities like CTS (Computed Tomography Scans) and MRI (Magnetic Resonance Imaging) with the aim of presenting better clinical content to clinicians and doctors for planning treatments or therapies. Prior studies based on white images have showed issues in early predictions of brain tumours like inaccurate image. This study attempts to overcome this issue by comparing MUNets (Modified-UNets), MCNNs (Multi-Cascaded ConvolutionNeural Networks) with fully connected CRFs (Conditional Random Fields), MFCLs (Modified Fully Connected Layers), TA-cGANs (Tissue-Aware conditional Generative Adversarial Networks) and EL (Ensemble Learning) algorithms. Furthermore, in order to enhance image quality and enable the early diagnosis of brain tumours, novel multimodal medical image fusion techniques are being examined. Furthermore, it is suggested that the EL algorithm improve MRI brain image fusion performance. The four primary phases of the suggested system are segmentation, image synthesis, feature extraction, and noise reduction. To eliminate noises, AMFs (Adaptive Median Filters) are used for reducing noises in MRI images and thus assist in enhancing classification accuracies. These features are taken into segmentation process using RGKMC (Region Growing based K-Means Clustering). Feature extractions are performed using AFFOCNNs (Adaptive FireFly Optimization based Convolution Neural Networks) algorithm which computes necessary and prominent features. Subsequently, MCCNNs, MUNets, MFCLs, TAcGANs and EL algorithms are applied for classifications through training and testing models. They classify features more accurately using their informative features. The experimental result proves that ELalgorithms provides better classificationsthan the MCCNNs, MUNets, MFCLs and TA-cGANs when evaluated in terms of higher accuracies, precisions, recalls and reduced MSEs (Mean Square Errors), execution times