Multi-class probabilistic atlas-based whole heart segmentation method in cardiac CT and MRI

• DOI: 10.48550/arxiv.2102.01822
• Publication Date: 2021-01-01
• Authors: Tarun Kanti Ghosh; Md. Kamrul Hasan; Shidhartho Roy; Md. Ashraful Alam; Eklas Hossain; Mohiuddin Ahmad
• Keywords: Margin (machine learning)

Accurate and robust whole heart substructure segmentation is crucial in developing clinical applications, such as computer-aided diagnosis surgery. However, of different substructures challenging because inadequate edge or boundary information, the complexity background texture, diversity substructures' sizes shapes. This article proposes a framework for multi-class employing non-rigid registration-based probabilistic atlas incorporating Bayesian framework. We also propose registration pipeline utilizing multi-resolution strategy obtaining highest attainable mutual information between moving fixed images. further incorporate into expectation-maximization algorithm implement deep convolutional neural network-based encoder-decoder networks ablation studies. All extensive experiments are conducted publicly available dataset containing 20 MRI CT cardiac The proposed approach exhibits an encouraging achievement, yielding mean volume overlapping error 14.5 % scans exceeding state-of-the-art results by margin 1.3 terms same metric. As provides better-results to delineate heart, it can be medical diagnostic aiding tool helping experts with quicker more accurate results.