Heart Chamber Segmentation from CT Using Convolutional Neural Networks.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiotherapy Planning

Background:

• Computed Tomography (CT) is crucial for radiotherapy planning, requiring accurate segmentation of organs and regions of interest.
• Existing cardiac segmentation methods primarily focus on the left ventricle, neglecting simultaneous segmentation of the entire heart.
• Accurate cardiac chamber segmentation is essential for precise radiotherapy and diagnostic evaluation.

Purpose of the Study:

• To develop and evaluate a novel convolutional neural network (CNN)-based method for simultaneous segmentation of all four cardiac chambers (left ventricle, right ventricle, left atrium, right atrium) in 3D CT images.
• To assess the accuracy and performance of the proposed deep learning model for cardiac chamber segmentation.
• To provide an automated tool for cardiac segmentation in the context of radiotherapy planning.

Main Methods:

• A 5-class convolutional neural network (CNN) model was designed for semantic segmentation.
• The model was trained and validated on 3D CT datasets.
• Segmentation included five categories: left ventricle, right ventricle, left atrium, right atrium, and background.

Main Results:

• The CNN-based method achieved an overall accuracy of 87.2% ± 3.3%.
• The model demonstrated an overall chamber accuracy of 85.6 ± 6.1% for segmenting all cardiac chambers.
• The results indicate robust performance in differentiating and segmenting individual cardiac chambers.

Conclusions:

• The developed deep learning approach enables accurate, simultaneous segmentation of all four cardiac chambers from 3D CT images.
• This automated segmentation method shows significant potential for improving cardiac segmentation in radiotherapy planning.
• The CNN-based technique offers a promising advancement for clinical applications requiring detailed cardiac imaging analysis.