Domain-specific pretraining of deep learning systems in gastrointestinal endoscopy improves performance over current state-of-the-art pretraining methods

Aims Pretraining might be more effective if the training data resemble the envisioned application. We investigated if pretraining on general endoscopic imagery results in a better performance of five existing AI systems with an application in gastro-intestinal endoscopy, compared to current state-of-the art pretraining approaches (i.e., supervised pretraining with ImageNet and semi-weakly supervised pretraining with the Billion-scale data set).

Methods Our group has created an endoscopy-specific dataset called GastroNet for pretraining deep learning systems in endoscopy. GastroNet consists of 5,084,494 endoscopic images retrospectively collected between 2012 and 2020 in seven Dutch hospitals. We created four pretrained models: one using GastroNet and three using the ImageNet and/or the Billion-scale data sets. The pretraining method was either supervised, self-supervised, or semi-weakly supervised. The pretrained models were subsequently trained towards five independent, commonly used applications in GI endoscopy, using their original application-specific datasets. The outcome parameters were classification and/or localization performance of the five trained applications

Results Results are presented in Table 1. Overall, the domain-specific pretrained model resulted in a statistically superior performance for the five different GI applications ([Fig. 1]).

Table 1

Conclusions Domain-specific pretraining is superior to current state-of-the-art pretraining approaches for developing deep learning algorithms in GI endoscopy. It also allows more effective use of the generally scarce application-specific endoscopy images. These findings might cause a paradigm shift in the development of AI systems in endoscopy.

Publication History

Article published online:
14 April 2023