EXAMINING INTEROBSERVER VARIABILITY IN HISTOPATHOLOGY REPORTS FOR LUNG ADENOCARCINOMA
Abstract
The difficulty in visually interpreting complex lung tumor patterns in histopathology images results in interobserver variability among pathologists, leading to diagnosis and treatment delays. This can be avoided by implementing a digital system that mediates disagreements by generating informative captions for pathology images based on feature extraction and pattern recognition. Extracting key diagnostic terms from historical pathology reports is the preliminary step to developing this system. This paper proposes an algorithm that generates relevant diagnostic terms from a pathology report based on a short list of known keywords using cosine similarity.