Self-supervised Disentanglement of Disease Effects from Aging in 3D Medical Shapes

Dimensions overall score 8.0

GitHub Code Pulse

No public code linked for this paper yet.

• The proposed two-stage framework combines unsupervised disease discovery with self-supervised disentanglement of implicit shape representations to separate pathological changes from physiological aging in 3D medical shapes.method 95%
  Evidencepartial

  To address this challenge, we propose a two-stage framework combining unsupervised disease discovery with self-supervised disentanglement of implicit shape representations.

  Implicationpartial

  Directly stated in abstract as the core methodological approach

  Verificationpartial

  partial

• The method achieves near-supervised performance on ADNI hippocampus and OAI distal femur shapes, improving disentanglement and reconstruction over state-of-the-art unsupervised baselines.result 90%
  Evidencepartial

  On ADNI hippocampus and OAI distal femur shapes, we achieve near-supervised performance, improving disentanglement and reconstruction over state-of-the-art unsupervised baselines.

  Implicationpartial

  Direct performance claim with specific datasets mentioned

  Verificationpartial

  partial

• The framework enables high-fidelity reconstruction, controllable synthesis, and factor-based explainability of 3D medical shapes.technical 85%
  Evidencepartial

  while enabling high-fidelity reconstruction, controllable synthesis, and factor-based explainability.

  Implicationpartial

  Directly stated capability of the method

  Verificationpartial

  partial

• The method uses pseudo disease labels discovered through clustering on shape latent space without ground-truth diagnosis during discovery.method 90%
  Evidencepartial

  We then apply clustering on the shape latent space, which yields pseudo disease labels without using ground-truth diagnosis during discovery.

  Implicationpartial

  Explicit description of the unsupervised labeling approach

  Verificationpartial

  partial

• The disentanglement is enforced using a multi-objective loss combining covariance and supervised contrastive loss with pseudo disease labels and ground truth age labels.technical 88%
  Evidencepartial

  We enforce separation and controllability with a multi-objective disentanglement loss combining covariance and a supervised contrastive loss.

  Implicationpartial

  Specific technical detail about the loss function

  Verificationpartial

  partial

• The method requires high-quality 3D medical imaging data which may not be available in all clinical settings.limitation 80%
  Evidencepartial

  Requires high-quality 3D medical imaging data which may not be available in all clinical settings

  Implicationpartial

  Stated as a caveat in the analysis section

  Verificationpartial

  partial

• Model performance depends on the quality of unsupervised clustering which could introduce biases.limitation 75%
  Evidencepartial

  Model performance depends on the quality of unsupervised clustering which could introduce biases

  Implicationpartial

  Explicit limitation mentioned in analysis

  Verificationpartial

  partial

• The approach is particularly valuable for pharmaceutical companies to improve patient selection for drug trials by identifying disease-specific biomarkers.market 70%
  Evidencepartial

  Pharmaceutical companies and clinical research organizations would pay for this product to improve patient selection for drug trials by identifying disease-specific biomarkers

  Implicationpartial

  Implied commercial application from analysis section

  Verificationpartial

  partial