Modeling Disagreement in Automatic Data Labelling for Semi-Supervised Learning in Clinical Natural Language Processing

TL;DR

This paper evaluates the uncertainty estimation capabilities of various models in clinical NLP, showing Gaussian Processes outperform others in quantifying risks in radiology report analysis.

Contribution

It introduces a comparative analysis of uncertainty estimation methods in healthcare NLP, highlighting the effectiveness of Gaussian Processes for risk quantification.

Findings

Gaussian Processes outperform other models in uncertainty quantification.

GPs provide better risk estimates with strong predictive performance.

Uncertainty estimation improves decision-making in clinical NLP applications.

Abstract

Computational models providing accurate estimates of their uncertainty are crucial for risk management associated with decision making in healthcare contexts. This is especially true since many state-of-the-art systems are trained using the data which has been labelled automatically (self-supervised mode) and tend to overfit. In this work, we investigate the quality of uncertainty estimates from a range of current state-of-the-art predictive models applied to the problem of observation detection in radiology reports. This problem remains understudied for Natural Language Processing in the healthcare domain. We demonstrate that Gaussian Processes (GPs) provide superior performance in quantifying the risks of 3 uncertainty labels based on the negative log predictive probability (NLPP) evaluation metric and mean maximum predicted confidence levels (MMPCL), whilst retaining strong predictive performance.