[Paper Review] A Novel Deep Learning Based Approach for Left Ventricle Segmentation in Echocardiography: MFP-Unet

Shakiba Moradi, Mostafa Ghelich Oghli+6 more|arXiv (Cornell University)|Jun 25, 2019

Cardiac Valve Diseases and Treatments59 references156 citations

TL;DR

This paper proposes MFP-Unet, a novel U-Net variant that enhances left ventricle (LV) segmentation in 2D echocardiography by fusing multi-level decoder feature maps through depth equalization and upsampling before final semantic segmentation. The method achieves state-of-the-art performance with a Dice Score of 0.945, Hausdorff Distance of 1.62, and strong clinical agreement on functional parameters.

ABSTRACT

Segmentation of the Left ventricle (LV) is a crucial step for quantitative measurements such as area, volume, and ejection fraction. However, the automatic LV segmentation in 2D echocardiographic images is a challenging task due to ill-defined borders, and operator dependence issues (insufficient reproducibility). U-net, which is a well-known architecture in medical image segmentation, addressed this problem through an encoder-decoder path. Despite outstanding overall performance, U-net ignores the contribution of all semantic strengths in the segmentation procedure. In the present study, we have proposed a novel architecture to tackle this drawback. Feature maps in all levels of the decoder path of U-net are concatenated, their depths are equalized, and up-sampled to a fixed dimension. This stack of feature maps would be the input of the semantic segmentation layer. The proposed network yielded state-of-the-art results when comparing with results from U-net, dilated U-net, and deeplabv3, using the same dataset. An average Dice Metric (DM) of 0.945, Hausdorff Distance (HD) of 1.62, Jaccard Coefficient (JC) of 0.97, and Mean Absolute Distance (MAD) of 1.32 are achieved. The correlation graph, bland-altman analysis, and box plot showed a great agreement between automatic and manually calculated volume, area, and length.

Motivation & Objective

• To address the challenge of poor reproducibility and inter-observer variability in manual LV segmentation of echocardiographic images.
• To improve upon U-Net's limitations by incorporating semantic strengths from all decoder levels.
• To develop a deep learning model that enhances segmentation accuracy in low-contrast, speckled echocardiographic images with ill-defined borders.
• To achieve clinically reliable segmentation results that match manual delineation in volume, ejection fraction, and area measurements.

Proposed method

• The proposed MFP-Unet architecture extends U-Net by concatenating feature maps from all decoder levels after depth equalization and upsampling to a fixed dimension.
• Multi-scale feature pyramids are constructed using dilated convolutions with a dilation rate of 2 to expand receptive fields without increasing parameters.
• A preprocessed input layer using Niblack’s local thresholding with k=2 is added to enhance edge contrast and improve feature learning.
• Skip connections are preserved between encoder and decoder paths to maintain spatial resolution and gradient flow.
• The final semantic segmentation layer processes the stacked, multi-level feature maps to produce precise LV boundary predictions.
• The network is trained end-to-end using supervised learning on a dataset of 2D 4-chamber echocardiographic views with manual ground truth.

Experimental results

Research questions

• RQ1 Can fusing multi-level decoder features improve LV segmentation accuracy in challenging echocardiographic images?
• RQ2 Does incorporating dilated convolutions with low dilation rates enhance feature representation without increasing model complexity?
• RQ3 How does MFP-Unet compare to U-Net, dilated U-Net, and DeepLabv3 in terms of segmentation metrics and clinical parameter consistency?
• RQ4 To what extent does the proposed method reduce inter-observer variability in LV functional parameter estimation?
• RQ5 Can the model achieve high accuracy with a relatively small, diverse training set?

Key findings

• MFP-Unet achieved a Dice Score of 0.945, indicating high overlap with manual segmentation.
• The Hausdorff Distance was 1.62, reflecting minimal maximum boundary error.
• The Jaccard Coefficient reached 0.97, confirming strong segmentation consistency.
• Mean Absolute Distance was 1.32, indicating high precision in boundary localization.
• Bland-Altman and correlation analyses showed negligible bias and strong agreement between automatic and manual measurements of end-diastolic volume, end-systolic volume, and ejection fraction.
• The boxplot analysis confirmed MFP-Unet's superiority in estimating ejection fraction with the narrowest interquartile range and minimal outliers.