MonoUNet: A Robust Tiny Neural Network for Automated Knee Cartilage Segmentation on Point-of-Care Ultrasound Devices

2026-04-09

AI Summary

This study addresses the challenges of substantial image appearance variability and limited computational resources in knee cartilage segmentation for portable ultrasound devices. To this end, the authors propose MonoUNet, an ultra-compact U-Net architecture that, for the first time, integrates trainable local phase (monogenic) features into an extremely lightweight network, augmented with a gated feature injection mechanism. This approach significantly enhances cross-device generalization and robustness while drastically reducing model complexity. Experimental results demonstrate Dice scores of 92.62%–94.82% and mean absolute symmetric surface distances (MASD) of 0.133–0.254 mm. Moreover, the model achieves a 10- to 700-fold reduction in parameters and a 14- to 2000-fold decrease in computational cost, with cartilage thickness and echogenicity measurements showing high agreement with manual annotations.

Abstract

Objective: To develop a robust and compact deep learning model for automated knee cartilage segmentation on point-of-care ultrasound (POCUS) devices. Methods: We propose MonoUNet, an ultra-compact U-Net consisting of (i) an aggressively reduced backbone with an asymmetric decoder, (ii) a trainable monogenic block that extracts multi-scale local phase features, and (iii) a gated feature injection mechanism that integrates these features into the encoder stages to reduce sensitivity to variations in ultrasound image appearance and improve robustness across devices. MonoUNet was evaluated on a multi-site, multi-device knee cartilage ultrasound dataset acquired using cart-based, portable, and handheld POCUS devices. Results: Overall, MonoUNet outperformed existing lightweight segmentation models, with average Dice scores ranging from 92.62% to 94.82% and mean average surface distance (MASD) values between 0.133 mm and 0.254 mm. MonoUNet reduces the number of parameters by 10x--700x and computational cost by 14x--2000x relative to existing lightweight models. MonoUNet cartilage outcomes showed excellent reliability and agreement with the manual outcomes: intraclass correlation coefficients (ICC$_{2,k})$=0.96 and bias=2.00% (0.047 mm) for average thickness, and ICC$_{2,k}$=0.99 and bias=0.80% (0.328 a.u.) for echo intensity. Conclusion: Incorporating trainable local phase features improves the robustness of highly compact neural networks for knee cartilage segmentation across varying acquisition settings and could support scalable ultrasound-based assessment and monitoring of knee osteoarthritis using POCUS devices. The code is publicly available at https://github.com/alvinkimbowa/monounet.

Problem

• knee cartilage segmentation
• point-of-care ultrasound
• robustness
• ultrasound image variability
• compact neural network

Innovation

• MonoUNet
• monogenic features
• ultra-compact neural network
• point-of-care ultrasound
• knee cartilage segmentation