M-Net: An encoder-decoder architecture for medical image analysis using ensemble learning

Only a few of the many subfields of biomedical science study include biomedical engineering, biomedical signal processing, gene analysis, and biomedical image processing. For the investigation and diagnosis of diseases, classification, detection, and recognition have tremendous importance. This work presents a fully automated deep-ensemble architecture, M-Net, for pixel-level semantic segmentation and classification of medical images. The performance of M-Net is evaluated by implementing it on the brain structural Magnetic Resonance Imaging (sMRI) for diagnosing Alzheimer's disease from various sources of datasets. The M-Net system successfully segmented the hippocampus region, vulnerable to damage at the early stage of AD, from the brain sMRI data. The obtained overall accuracy of 99% shows that the proposed deep learning technique is superior to the existing deep semantic segmentation techniques and can reduce the diagnostic time of radiologists.