NEMESIS: Noise-suppressed Efficient MAE with Enhanced Superpatch Integration Strategy

Dimensions overall score 7.0

• NEMESIS achieves a mean AUROC of 0.9633 on the BTCV multi-organ classification benchmark with a frozen backbone and linear classifier, surpassing fully fine-tuned SuPreM (0.9493) and VoCo (0.9387).result 95%
  Evidencepartial

  On the BTCV multi-organ classification benchmark, NEMESIS with a frozen backbone and a linear classifier achieves a mean AUROC of 0.9633, surpassing fully fine-tuned SuPreM (0.9493) and VoCo (0.9387).

  Implicationpartial

  Directly stated in abstract with specific numeric results comparing against named baselines

  Verificationpartial

  partial

• NEMESIS retains an AUROC of 0.9075 under a low-label regime with only 10% of available annotations, demonstrating strong label efficiency.result 95%
  Evidencepartial

  Under a low-label regime with only 10% of available annotations, it retains an AUROC of 0.9075, demonstrating strong label efficiency.

  Implicationpartial

  Directly stated in abstract with specific numeric result for low-label regime

  Verificationpartial

  partial

• NEMESIS reduces computational cost to 31.0 GFLOPs per forward pass compared to 985.8 GFLOPs for the full-volume baseline.technical 95%
  Evidencepartial

  Furthermore, the superpatch-based design reduces computational cost to 31.0 GFLOPs per forward pass, compared to 985.8 GFLOPs for the full-volume baseline

  Implicationpartial

  Directly stated in abstract with specific numeric comparison of computational costs

  Verificationpartial

  partial

• NEMESIS introduces Masked Anatomical Transformer Blocks (MATB) that perform dual-masking through parallel plane-wise and axis-wise token removal.method 90%
  Evidencepartial

  Masked Anatomical Transformer Blocks (MATB) that perform dual-masking through parallel plane-wise and axis-wise token removal

  Implicationpartial

  Directly stated as a key component of the method in the abstract

  Verificationpartial

  partial

• NEMESIS introduces NEMESIS Tokens (NT) for cross-scale context aggregation.method 90%
  Evidencepartial

  NEMESIS Tokens (NT) for cross-scale context aggregation

  Implicationpartial

  Directly stated as a key component of the method in the abstract

  Verificationpartial

  partial

• NEMESIS uses noise-enhanced reconstruction as a pretext task.method 90%
  Evidencepartial

  noise-enhanced reconstruction as a pretext task

  Implicationpartial

  Directly stated as a key component of the method in the abstract

  Verificationpartial

  partial

• NEMESIS operates on local 128x128x128 superpatches to enable memory-efficient training while preserving anatomical detail.method 90%
  Evidencepartial

  NEMESIS, a masked autoencoder (MAE) framework that operates on local 128x128x128 superpatches, enabling memory-efficient training while preserving anatomical detail

  Implicationpartial

  Directly stated in abstract describing the core architectural approach

  Verificationpartial

  partial

• Applying SSL to 3D CT remains challenging due to the high memory cost of full-volume transformers and the anisotropic spatial structure of CT data.limitation 85%
  Evidencepartial

  applying SSL to 3D CT remains challenging due to the high memory cost of full-volume transformers and the anisotropic spatial structure of CT data

  Implicationpartial

  Directly stated as motivation for the work, though presented as a general challenge rather than a specific limitation of NEMESIS

  Verificationpartial

  partial

Competitive landscape

A memory-efficient self-supervised learning framework for 3D medical imaging that significantly reduces computational cost and improves performance with limited annotations.