Traditional radiology reports typically include the patient's symptoms, imaging findings, and final diagnosis, but they often lack a detailed description of the causal relationships and reasoning that led to the diagnosis. Therefore, there is a need for information about the process of how a diagnosis is made. Moreover, understanding causality is important for studying the decision-making process required to make a diagnosis.
The Hidden-Rad Challenge aims to develop the ability to explain why a diagnosis is made during the image reading process as an introduction to producing accurate and meaningful medical reports. The goal of the challenge is to assess how well a diagnosis or diagnostic process can be included in relation to a given radiology image or medical report in the area of MIMIC database.
Participants are required to provide their findings within a given segment. There are two subtasks.
Task 1: Generate a report that provides diagnostic inferences by identifying hidden causalities in MIMIC-CXR reports.
The MIMIC report and DICOM images of each case are analyzed by AI to create a causality report, which is then validated against the correct answer data. The goal is to produce a causality report that reflects the way a medical professional would diagnose.
Task 2: Simulate a radiologist's decision-making process to generate a report from initial impression to final conclusion.
Using crowdsourced data from each case's first impression (A1) to final impression (A4), generate a report with causal information and validate it against the correct answer data.
The dataset used is based on the MIMIC dataset, a medical dataset jointly developed and provided by the Massachusetts Institute of Technology (MIT) and Beth Israel Deaconess Medical Center (BIDMC). Konyang University, in these tasks, provides crowdsourced results of real radiologists' readings based on the MIMIC dataset.
The challenge will build on this foundation, inviting participants to push the boundaries of medical report reconstruction, enhancing not only information retrieval and summarization but also causal reasoning in diagnostic contexts.
This challenge emphasizes the development of AI systems capable of understanding and generating more comprehensive medical reports that reflect the nuanced decision-making processes of human radiologists.