← All posts

EAT

E-AT for calibrated medical-image classification · macro-F1 0.8647 · ECE 1.03% vs. CR-SAM 1.7%

Built with PyTorch · ConvNeXtV2-Tiny · Focal loss · R-Drop · FGM · ISIC 2019

E-AT result summary on balanced four-class ISIC: macro-F1 0.8647 and minimum expected calibration error 1.03 percent at separate checkpoints
Balanced four-class ISIC · ConvNeXtV2-Tiny 384 px

WHY

In medical image classification, a correct label is not enough if the confidence cannot be trusted. Two models with similar accuracy can create very different risk when one remains overconfident on its mistakes. The research therefore asked whether classification quality and confidence calibration could be improved within the same training objective.

ISIC 4 classes

Balanced · ConvNeXtV2-Tiny · 384 px

Macro-F1 0.8647

Best classification result

Minimum ECE 1.03%

CR-SAM (AAAI 2024): 1.7%

HOW

One objective for classification and calibrated confidence

E-AT couples sample difficulty, perturbation, and prediction consistency instead of calibrating only after training.

  1. 01 Image and class label

    Run a clean stochastic forward pass and measure target-class confidence on each ISIC image.

  2. 02 Focal difficulty

    Give difficult samples more weight through focal loss rather than treating every example equally.

  3. 03 Adaptive FGM

    Scale a bounded perturbation with sample difficulty and evaluate a second stochastic prediction.

  4. 04 R-Drop update

    Optimize classification and bidirectional prediction consistency together in one update.

RESULT

ISIC result

MetricE-AT resultSelection
Macro-F1 ↑0.8647Classification checkpoint
ECE ↓1.03%Calibration checkpoint
CR-SAM ECE ↓1.7%AAAI 2024 reference

Macro-F1 and ECE report their respective best checkpoints.

E-AT training diagram linking focal sample difficulty, adaptive FGM perturbation, and bidirectional R-Drop consistency
Focal loss, adaptive FGM, and bidirectional R-Drop are optimized together

Contribution

  • Designed the E-AT objective that couples focal difficulty, adaptive FGM, and bidirectional R-Drop.
  • Built the balanced ISIC training and evaluation pipeline and tracked macro-F1 and ECE separately.

Evidence

Private project