IVCVApr 21, 2021

Rethinking Annotation Granularity for Overcoming Shortcuts in Deep Learning-based Radiograph Diagnosis: A Multicenter Study

arXiv:2104.10553v49 citations
Originality Incremental advance
AI Analysis

This addresses the issue of poor generalizability in medical imaging AI for radiologists and clinicians, offering a practical solution to improve diagnostic accuracy across datasets.

The study tackled the problem of shortcut learning in deep learning-based radiograph diagnosis by comparing models trained with radiograph-level vs. lesion-level annotations, finding that fine-grained annotations significantly improved external classification and lesion detection performance, e.g., CheXDet achieved AUCs of 0.67 vs. 0.51 for fracture on NIH-Google and JAFROC-FOMs of 0.87 vs. 0.13 for pneumothorax on the internal set.

Two DL models were developed using radiograph-level annotations (yes or no disease) and fine-grained lesion-level annotations (lesion bounding boxes), respectively named CheXNet and CheXDet. The models' internal classification performance and lesion localization performance were compared on a testing set (n=2,922), external classification performance was compared on NIH-Google (n=4,376) and PadChest (n=24,536) datasets, and external lesion localization performance was compared on NIH-ChestX-ray14 dataset (n=880). The models were also compared to radiologists on a subset of the internal testing set (n=496). Given sufficient training data, both models performed comparably to radiologists. CheXDet achieved significant improvement for external classification, such as in classifying fracture on NIH-Google (CheXDet area under the ROC curve [AUC]: 0.67, CheXNet AUC: 0.51; p<.001) and PadChest (CheXDet AUC: 0.78, CheXNet AUC: 0.55; p<.001). CheXDet achieved higher lesion detection performance than CheXNet for most abnormalities on all datasets, such as in detecting pneumothorax on the internal set (CheXDet jacknife alternative free-response ROC-figure of merit [JAFROC-FOM]: 0.87, CheXNet JAFROC-FOM: 0.13; p<.001) and NIH-ChestX-ray14 (CheXDet JAFROC-FOM: 0.55, CheXNet JAFROC-FOM: 0.04; p<.001). To summarize, fine-grained annotations overcame shortcut learning and enabled DL models to identify correct lesion patterns, improving the models' generalizability.

Foundations

The foundational work for this paper's niche, ranked by how specifically the neighbourhood builds on it — not by global fame.

Your Notes