Causal Analysis and Adversarial Methods in Medical Diagnostics

Innovative Research in Medical Diagnostics

Exploring causal reasoning and adversarial methods to enhance medical diagnostic models through rigorous analysis and experimental validation.

A series of Scrabble tiles spelling out the word 'DIAGNOSIS' is arranged horizontally on a surface. The background is divided, with a dark upper section and a light lower section.

Causal Analysis

This project explores causal fallacies in medical diagnostic models.

A person in a medical setting is lying on a table, about to enter a large cylindrical medical imaging machine labeled 'Radixact'. Another person, wearing a white coat, stands beside the machine, using a touchscreen panel.

Adversarial Method

We propose a new adversarial sample generation method for exposing causal fallacies in medical models through theoretical analysis and experimental validation using public datasets and simulated environments.

A medical imaging device with a monitor displaying a user interface featuring multiple icons and buttons. The device is labeled 3D OCT Maestro by Topcon. There is also a black adjustable arm and other medical supplies in the background.

Comparative Study

The project includes comparative experiments to evaluate the effectiveness of our method against traditional approaches in optimizing model performance and exposing fallacies in diagnostic models.