University of Twente Student Theses

Login

Comparative Analysis and Optimization of Model-Method Combinations for Out-of-Distribution Detection in Medical Image Classification

Cho, Gyum (2024) Comparative Analysis and Optimization of Model-Method Combinations for Out-of-Distribution Detection in Medical Image Classification.

[img] PDF
794kB
Abstract:AI is now replacing humans in various fields previously exclusive to humanity. Medical testing, scanning evaluation, and categorizing symptoms were sensitive tasks where human errors could lead to fatal situations. AI’s high performance and low risk of missing critical factors have rapidly replaced this area. However, even with a high success rate, medical AI shows serious abnormal behavior when it faces new input far from the training domain. This problem has been classified as an out-of-distribution (OOD) problem. This paper analyzes the main factors of abnormal behavior when AI faces OOD problems. The external factors influencing this abnormal behavior will be examined and regulated with statistical observation throughout the training process. By incorporating evidence from experiments, this research demonstrates an approach to the OOD problem using three distinct pre-trained AI models. Each AI model will be trained with a medical MNIST dataset, and the statistical evaluation will prove which factors mainly affect the abnormal behavior of the AI. The result can prevent an OOD problem when AI classification models are used. Both patients and the collecting organization have consented to all medical images used for AI training.
Item Type:Essay (Bachelor)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:31 mathematics
Programme:Computer Science BSc (56964)
Link to this item:https://purl.utwente.nl/essays/100997
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page