Semi-supervised classification of radiology images with NoTeacher: A Teacher that is not Mean

TL;DR

NoTeacher is a novel semi-supervised learning framework for radiology image classification that eliminates the need for a teacher network, achieving high accuracy with minimal labeled data across various imaging modalities.

Contribution

Introduces NoTeacher, a consistency-based SSL method with independent networks, adaptable to diverse radiology imaging challenges, outperforming existing SSL methods with minimal tuning.

Findings

Achieves 90-95% of fully supervised AUROC with less than 15% labeled data.

Outperforms established SSL methods in radiology image classification.

Effective across 2D/3D inputs and multi-label scenarios.

Abstract

Deep learning models achieve strong performance for radiology image classification, but their practical application is bottlenecked by the need for large labeled training datasets. Semi-supervised learning (SSL) approaches leverage small labeled datasets alongside larger unlabeled datasets and offer potential for reducing labeling cost. In this work, we introduce NoTeacher, a novel consistency-based SSL framework which incorporates probabilistic graphical models. Unlike Mean Teacher which maintains a teacher network updated via a temporal ensemble, NoTeacher employs two independent networks, thereby eliminating the need for a teacher network. We demonstrate how NoTeacher can be customized to handle a range of challenges in radiology image classification. Specifically, we describe adaptations for scenarios with 2D and 3D inputs, uni and multi-label classification, and class distribution mismatch between labeled and unlabeled portions of the training data. In realistic empirical evaluations on three public benchmark datasets spanning the workhorse modalities of radiology (X-Ray, CT, MRI), we show that NoTeacher achieves over 90-95% of the fully supervised AUROC with less than 5-15% labeling budget. Further, NoTeacher outperforms established SSL methods with minimal hyperparameter tuning, and has implications as a principled and practical option for semisupervised learning in radiology applications.