FULLY AUTOMATIC SEGMENTATION OF THE RIGHT VENTRICLE VIA MULTI-TASK DEEP NEURAL NETWORKS

Area of Science:

• Medical imaging analysis
• Cardiovascular imaging
• Artificial intelligence in medicine

Background:

• Cardiac magnetic resonance (MR) imaging is crucial for diagnosing heart conditions.
• Accurate segmentation of cardiac ventricles, particularly the right ventricle (RV), is essential for clinical parameter extraction and patient prognosis.
• Existing fully convolutional network (FCN) methods for RV segmentation have limitations.

Purpose of the Study:

• To propose a novel multi-task deep neural network (DNN) architecture for enhanced automatic right ventricle (RV) segmentation from cardiac MR images.
• To leverage shared features across tasks for improved segmentation performance.
• To provide a more accurate tool for cardiac pathology prognosis.

Main Methods:

• Development and implementation of a multi-task U-net architecture using the Tensorflow framework.
• The proposed DNN can utilize any FCN as a base.
• The model is designed for efficient end-to-end training.

Main Results:

• The multi-task DNN demonstrated improved segmentation performance compared to existing methods.
• The approach showed particular effectiveness in segmenting small-sized right ventricles (RVs).
• Numerical tests on real cardiac MR datasets validated the proposed method's capabilities.

Conclusions:

• The proposed multi-task DNN architecture offers a significant advancement in automatic RV segmentation from cardiac MR images.
• This enhanced segmentation accuracy, especially for small RVs, can lead to better clinical parameter assessment and improved cardiac pathology prognosis.
• The flexible multi-task framework allows for integration with various FCNs, paving the way for future developments in cardiovascular image analysis.