Showing 1–1 of 1 results for author: H, M P

Extracting and Learning Fine-Grained Labels from Chest Radiographs

Authors: Tanveer Syeda-Mahmood, Ph. D, K. C. L Wong, Ph. D, Joy T. Wu, M. D., M. P. H, Ashutosh Jadhav, Ph. D, Orest Boyko, M. D. Ph. D

Abstract: Chest radiographs are the most common diagnostic exam in emergency rooms and intensive care units today. Recently, a number of researchers have begun working on large chest X-ray datasets to develop deep learning models for recognition of a handful of coarse finding classes such as opacities, masses and nodules. In this paper, we focus on extracting and learning fine-grained labels for chest X-ray… ▽ More Chest radiographs are the most common diagnostic exam in emergency rooms and intensive care units today. Recently, a number of researchers have begun working on large chest X-ray datasets to develop deep learning models for recognition of a handful of coarse finding classes such as opacities, masses and nodules. In this paper, we focus on extracting and learning fine-grained labels for chest X-ray images. Specifically we develop a new method of extracting fine-grained labels from radiology reports by combining vocabulary-driven concept extraction with phrasal grou** in dependency parse trees for association of modifiers with findings. A total of 457 fine-grained labels depicting the largest spectrum of findings to date were selected and sufficiently large datasets acquired to train a new deep learning model designed for fine-grained classification. We show results that indicate a highly accurate label extraction process and a reliable learning of fine-grained labels. The resulting network, to our knowledge, is the first to recognize fine-grained descriptions of findings in images covering over nine modifiers including laterality, location, severity, size and appearance. △ Less

Submitted 18 November, 2020; originally announced November 2020.

Comments: This paper won the Homer R. Warner Award at AMIA 2020 awarded to a paper that best describes approaches to improving computerized information acquisition, knowledge data acquisition and management, and experimental results documenting the value of these approaches. The paper shows a combination of textual and visual processing to automatically recognize complex findings in chest X-rays