Truace
Subscribe

New AI method helps diagnose infant hip dysplasia from ultrasound with few training images

Study title: Dual-domain cross-constraint MIM based hybrid symmetry convolution-Mamba network for ultrasound diagnosis of DDH

Researchers proposed DCM-SCMNet, a network for diagnosing developmental dysplasia of the hip in infants from B-mode ultrasound images. It targets the small-sample problem, in which too few labeled images are available to train deep learning models well.

Read the study abstract

Developmental dysplasia of the hip (DDH) is a critical congenital condition in infants, which is generally early diagnosed by B-mode ultrasound (BUS). Deep learning-based computer-aided diagnosis has shown effectiveness in diagnosing DDH from BUS images. However, the issue of small sample size remains a challenge. To this end, a Dual-domain Cross-constraint Masked image modeling based hybrid Symmetry Convolution-Mamba Network (DCM-SCMNet) is proposed for the BUS-based diagnosis of DDH with limited training sampl…

By the Truvace desk1 min read1 sourceEvidence: Moderate · 52/100 — Benchmark or lab evaluation · 1 independent sourcePeer-reviewedTRV-2026-2260· cite & verify
Dual-domain cross-constraint MIM based hybrid symmetry convolution-Mamba network for ultrasound diagnosis of DDH"Medical Laboratory" by ben.dracup, CC BY 2.0.

In brief

The approach matters because early diagnosis of this congenital condition relies on ultrasound, and a model that works with limited data could make computer-aided diagnosis more practical. The abstract reports only comparative performance on one dataset, so real-world clinical benefit remains unshown.

Main points

  1. The model is designed for diagnosing developmental dysplasia of the hip from B-mode ultrasound images when training samples are limited.
  2. Pretraining uses masked image modeling in both the spatial and frequency domains, with a shared hybrid encoder and two decoders.
  3. A dual-domain cross-constraint strategy links the two decoders' reconstructions to help learn more inherent feature representations.
  4. The encoder combines symmetry convolution with a symmetry Mamba block that scans bidirectionally to capture the anatomical symmetry of hip bone structure.

The gain

A new deep learning model trained with masked image modeling on limited ultrasound samples diagnosed infant hip dysplasia better than the other algorithms it was compared with on the DDH dataset.

The rundown

The study starts from the observation that deep learning-based computer-aided diagnosis has already shown effectiveness in diagnosing hip dysplasia from ultrasound, but small sample size remains a challenge.

In the pretext task, a shared hybrid encoder and two decoders reconstruct masked images in spatial and frequency domains. The cross-constraint strategy is meant to improve feature learning from these reconstructions.

The novel encoder, SCMNet, integrates symmetry convolution and symmetry Mamba. The Mamba component uses a bidirectional scanning mechanism tailored to bone structure symmetry and is described as maintaining computational efficiency.

According to the abstract, experimental results show DCM-SCMNet outperforms all compared algorithms on the DDH dataset.

Sources

  1. Peer-reviewedMedical & Biological Engineering & Computing2026-10-24

The debate