A Novel Deep Learning Method for Segmenting the Left Ventricle in Cardiac Cine MRI

TL;DR

The paper addresses the challenge of accurate left-ventricle segmentation in short-axis cine MRI and proposes GBU-Net, a U-Net–based architecture augmented with batch-group normalization and ELU activation. By systematically evaluating normalization strategies and employing encoder optimization plus data augmentation, it achieves a mean dice score of about 0.97 on the Sunnybrook LV dataset, with training time per epoch reduced to 7 seconds. This approach outperforms standard U-Net variants, illustrating improved boundary delineation and contextual understanding essential for clinical analysis and potential surgical robotics applications. The work demonstrates the practical benefit of combining BN and GN in a unified normalization framework and highlights directions for real-time guidance and 3D reconstruction to further impact medical imaging workflows.

Abstract

This research aims to develop a novel deep learning network, GBU-Net, utilizing a group-batch-normalized U-Net framework, specifically designed for the precise semantic segmentation of the left ventricle in short-axis cine MRI scans. The methodology includes a down-sampling pathway for feature extraction and an up-sampling pathway for detail restoration, enhanced for medical imaging. Key modifications include techniques for better contextual understanding crucial in cardiac MRI segmentation. The dataset consists of 805 left ventricular MRI scans from 45 patients, with comparative analysis using established metrics such as the dice coefficient and mean perpendicular distance. GBU-Net significantly improves the accuracy of left ventricle segmentation in cine MRI scans. Its innovative design outperforms existing methods in tests, surpassing standard metrics like the dice coefficient and mean perpendicular distance. The approach is unique in its ability to capture contextual information, often missed in traditional CNN-based segmentation. An ensemble of the GBU-Net attains a 97% dice score on the SunnyBrook testing dataset. GBU-Net offers enhanced precision and contextual understanding in left ventricle segmentation for surgical robotics and medical analysis.

Figures (6)

• Figure 1: The BNU-Net convolutional network is architecturally composed of two primary components: (a) the contraction path designed for feature extraction, and (b) the systematic application of batch normalization following each convolutional operation within the convolutional layers.
• Figure 2: The architecture of the proposed fully convolutional network, designated as LNU-Net, integrates layer normalization processes following each convolutional operation within the convolutional layers.
• Figure 3: The GBU-Net, a fully convolutional network, is characterized by its unique architecture. It systematically incorporates both group normalization and batch normalization after each convolution within its convolutional layers.
• Figure 4: Architecture of the IBU-Net convolutional network. Instance normalization is applied in the first convolutional layer. Batch normalization is performed after each convolution in the convolutional layer.
• Figure 5: Overview of data rotation steps in the proposed sampling method