A Dense RNN for Sequential Four-Chamber View Left Ventricle Wall Segmentation and Cardiac State Estimation

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Research

Background:

• Accurate left ventricle (LV) segmentation is crucial for diagnosing cardiac diseases and studying cardiac mechanisms.
• Existing methods struggle with the complexity of four-chamber view cardiac images and diverse wall motion.
• Sequential segmentation of the LV wall in cardiac MRI is challenging due to intricate structures and motion variability.

Purpose of the Study:

• To develop a robust algorithm for accurate left ventricle (LV) wall segmentation in four-chamber view cardiac MRI time sequences.
• To improve the accuracy and efficiency of cardiac state estimation based on LV segmentation.
• To address the limitations of current deep learning approaches in handling sequential cardiac image data.

Main Methods:

• A dense recurrent neural network (RNN) incorporating Long Short-Term Memory (LSTM) cells is proposed.
• Two RNNs are combined: one for initial frame information and a second for segmentation generation.
• The dense RNN architecture enhances information flow between LSTM cells for improved sequential accuracy.

Main Results:

• The proposed dense RNN significantly improves frame-wise segmentation accuracy for cardiac MRI sequences.
• Achieved an Intersection over Union (IoU) of 92.13%, outperforming classical deep learning algorithms.
• Successfully enabled robust cardiac state estimation from segmented LV data.

Conclusions:

• The novel dense RNN method offers a stable and accurate solution for time-sequential LV segmentation in cardiac MRI.
• This approach represents a significant advancement in both cardiac image segmentation and state estimation.
• The method effectively utilizes cardiac sequence information, surpassing frame-by-frame analysis.