Unsupervised Anomaly Detection on Implicit Shape representations for Sarcopenia Detection

TL;DR

This work tackles sarcopenia detection by shifting from mass-based metrics to muscle shape analysis. It introduces an implicit neural representation (INR) with an autodecoder-based latent code to learn a shape prior from normal muscles and performs unsupervised anomaly detection through reconstruction errors and latent-space separation. On 103 segmented volumes from two in-house 3D ultrasound datasets, reconstruction Dice scores enable a practical anomaly threshold around $0.93$, and linear discriminant analysis of the latent codes clearly separates sarcopenic from normal muscles, suggesting clinical utility as a supplementary diagnostic aid. The approach offers a data-efficient, unsupervised pathway to quantify shape abnormalities and could be extended to other muscles and automated segmentation in future work.

Abstract

Sarcopenia is an age-related progressive loss of muscle mass and strength that significantly impacts daily life. A commonly studied criterion for characterizing the muscle mass has been the combination of 3D imaging and manual segmentations. In this paper, we instead study the muscles' shape. We rely on an implicit neural representation (INR) to model normal muscle shapes. We then introduce an unsupervised anomaly detection method to identify sarcopenic muscles based on the reconstruction error of the implicit model. Relying on a conditional INR with an auto-decoding strategy, we also learn a latent representation of the muscles that clearly separates normal from abnormal muscles in an unsupervised fashion. Experimental results on a dataset of 103 segmented volumes indicate that our double anomaly detection strategy effectively discriminates sarcopenic and non-sarcopenic muscles.

Figures (4)

• Figure 1: Overview of the training and testing phases.
• Figure 2: (left) Example muscles from the dataset. (right) Distribution of the muscle volumes for each fold, considering the train/test split vs. the average volume of RF muscle in patients with sarcopenia.
• Figure 3: (a) Volumetric error in % and (b) Dice Score by fold for the five test sets separated into "Test-healthy" (young and older adults, non-sarcopenic) and "Test-sarcopenia" (sarcopenic). (c) LDA applied on the train and test latents $z$ of fold 1.
• Figure 4: Qualitative results. Muscles from older adults with (a) and without (b) sarcopenia vs. from young subjects (c and d). Superposition of the GT (in green) and the prediction (in red).