Rule-based Automated Cancer Staging from Scanned Pathology Reports

Pathology reports provide detailed information regarding the patient's cancer diagnosis and stage, which are not only critical for designing effective treatment plans, but also serve as a critical resource for various clinical research endeavors. However, pathology reports are typically scanned into PDFs, which poses significant limitations on the machine's ability to extract informative features. Additionally, the effectiveness of data-driven models hinges on the quality and quantity of training data available. Pathology reports associated with a specific clinic may not meet the requirement to sufficiently train large-scale artificial intelligence models. To cope with those challenges, this paper presents a rulebased cancer staging tool designed to automatically identify cancer stages from scanned reports. We first develop an effective process to convert non-text-based PDF reports into a machine-interpretable format through the combination of image pre-processing and pretrained Optical Character Recognition (OCR). Second, we incorporate clinical rules for cancer staging, defined by the American Joint Committee on Cancer (AJCC), into the automated process, avoiding the requirement of large data for training the model. Finally, we build a user-friendly interface for our developed method, allowing clinicians to perform the end-to-end automated cancer staging directly from raw scanned reports.