PT  - JOURNAL ARTICLE
AU  - Olle G. Holmberg
AU  - Niklas D. Köhler
AU  - Thiago Martins
AU  - Jakob Siedlecki
AU  - Tina Herold
AU  - Leonie Keidel
AU  - Ben Asani
AU  - Johannes Schiefelbein
AU  - Siegfried Priglinger
AU  - Karsten U. Kortuem
AU  - Fabian J. Theis
TI  - Self-supervised retinal thickness prediction enables deep learning from unlabeled data to boost classification of diabetic retinopathy
AID  - 10.1101/861757
DP  - 2019 Jan 01
TA  - bioRxiv
PG  - 861757
4099  - http://biorxiv.org/content/early/2019/12/02/861757.short
4100  - http://biorxiv.org/content/early/2019/12/02/861757.full
AB  - Access to large, annotated samples represents a considerable challenge for training accurate deep-learning models in medical imaging. While current leading-edge transfer learning from pre-trained models can help with cases lacking data, it limits design choices, and generally results in the use of unnecessarily large models. We propose a novel, self-supervised training scheme for obtaining high-quality, pre-trained networks from unlabeled, cross-modal medical imaging data, which will allow for creating accurate and efficient models. We demonstrate this by accurately predicting optical coherence tomography (OCT)-based retinal thickness measurements from simple infrared (IR) fundus images. Subsequently, learned representations outperformed advanced classifiers on a separate diabetic retinopathy classification task in a scenario of scarce training data. Our cross-modal, three-staged scheme effectively replaced 26,343 diabetic retinopathy annotations with 1,009 semantic segmentations on OCT and reached the same classification accuracy using only 25% of fundus images, without any drawbacks, since OCT is not required for predictions. We expect this concept will also apply to other multimodal clinical data-imaging, health records, and genomics data, and be applicable to corresponding sample-starved learning problems.