Knowing When Not to Predict: Self Supervised Learning and Abstention for Safer DR Screening

Muskaan Chopra, Lorenz Sparrenberg, Jan H. Terheyden, Rafet Sifa

arXiv:2605.19133 · 2026-05-20 공개 · arXiv · PDF

self-supervised-learning diabetic-retinopathy calibrated-confidence selective-prediction medical-image-models pretraining-duration reliability-evaluation abstention

Abstract

Self-supervised learning (SSL) is now a standard way to pretrain medical image models, but performance is still mostly judged by downstream accuracy. For safety-critical screening tasks such as diabetic retinopathy grading, this is not enough: a model must also know when its predictions are unreliable and defer uncertain cases for clinical review. In this work, we examine how the length of SSL pretraining influences calibrated confidence and confidence-based abstention. We evaluate multiple SSL checkpoints under a fixed fine-tuning protocol and assess calibrated confidence, coverage, selective accuracy, and selective macro-F1. Across datasets and data regimes, SSL pretraining improves selective prediction compared to training from scratch. Unlike prior SSL studies that primarily evaluate downstream accuracy or AUROC, we analyze how SSL pretraining duration influences confidence behavior under calibrated confidence-based abstention. However, once accuracy saturates, selective performance can still change markedly across checkpoints, and longer pretraining does not consistently improve reliability. These results underscore the importance of abstention-aware evaluation and suggest that pretraining length should be treated as an important reliability-related design choice rather than only a computational detail. Code is available at GitHub.

한국어 요약

한 줄 요약

**[자기지도 학습 / 안전 임계 의료 AI]** 당뇨망막병증(DR) 스크리닝에서 SSL 사전학습 기간이 calibrated confidence와 신뢰도 기반 abstention 성능에 미치는 영향을 체계적으로 분석.

핵심 기여도

핵심 아이디어

"정확도와 신뢰도는 같은 곡선을 따르지 않는다"는 관찰. 특히 안전 임계 스크리닝에서는 모델이 자신의 불확실성을 적절히 표현하고 임상의 검토로 위임하는 능력이 정확도 자체보다 중요한 신뢰성 지표가 되어야 한다는 입장.

기술적 접근법

주요 결과

의의 및 한계

**의의**: SSL 사전학습 기간을 "연산 자원의 문제"가 아니라 reliability와 관련된 설계 결정으로 재정의함으로써 안전 임계 의료 AI 평가 프로토콜에 새로운 축을 도입. **한계**: 평가가 DR 스크리닝 중심이며, 다른 임상 태스크·모달리티로의 일반화는 후속 연구 과제.

실용적 활용