Medical Imaging Anomaly Detection: Dataset and Method Analysis

Medical anomaly detection is an important research area because abnormal medical images are often scarce, expensive to obtain, and require expert annotation. However, existing studies are scattered across different datasets, medical domains, evaluation protocols, and training settings, making fair comparison and reproducibility assessment difficult. This thesis presents a dataset-centric survey and experimental evaluation of medical anomaly detection methods. The survey covers 31 datasets from 11 medical domains and reviews 30 research papers. organized into five categories: four method families (supervised, unsupervised, VLM few-shot, VLM zero-shot) plus a set of benchmark and dataset papers. A dataset-method coverage analysis and a domain-wise gap analysis are also presented to provide a comprehensive overview of the current research landscape. To support the survey, representative methods that are MVFA, MediCLIP, MadCLIP are reproduced using their publicly available implementations from GitHub and evaluated. The reproduced AUC scores are then compared with those reported in the original papers to assess the reproducibility and analyze performance of methods across different medical domains. The experimental results show that the performance of the same method can vary across different datasets, which indicates that dataset characteristics influence anomaly detection performance. The four-shot evaluation setting also reveals performance differences that are less apparent under more less constrained evaluation settings like 16-, 24- and 32-shot. Overall, this work provides a unified analysis of datasets, methods, reproducibility, and evaluation practices in medical anomaly detection, and highlights the opportunities for future research and standardized benchmarking.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-15
DOI
https://doi.org/10.5281/zenodo.24257340
Primary Topic
Anomaly Detection Techniques and Applications
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Medical Imaging Anomaly Detection: Dataset and Method Analysis

Information Technology University Of Punjab, Ruqia,Ghulam
Zenodo (CERN European Organization for Nuclear Research)
Anomaly Detection Techniques and Applications
article

Medical Imaging Anomaly Detection: Dataset and Method Analysis

Information Technology University Of Punjab, Ruqia,Ghulam
article en

Abstract

Medical anomaly detection is an important research area because abnormal medical images are often scarce, expensive to obtain, and require expert annotation. However, existing studies are scattered across different datasets, medical domains, evaluation protocols, and training settings, making fair comparison and reproducibility assessment difficult. This thesis presents a dataset-centric survey and experimental evaluation of medical anomaly detection methods. The survey covers 31 datasets from 11 medical domains and reviews 30 research papers. organized into five categories: four method families (supervised, unsupervised, VLM few-shot, VLM zero-shot) plus a set of benchmark and dataset papers. A dataset-method coverage analysis and a domain-wise gap analysis are also presented to provide a comprehensive overview of the current research landscape. To support the survey, representative methods that are MVFA, MediCLIP, MadCLIP are reproduced using their publicly available implementations from GitHub and evaluated. The reproduced AUC scores are then compared with those reported in the original papers to assess the reproducibility and analyze performance of methods across different medical domains. The experimental results show that the performance of the same method can vary across different datasets, which indicates that dataset characteristics influence anomaly detection performance. The four-shot evaluation setting also reveals performance differences that are less apparent under more less constrained evaluation settings like 16-, 24- and 32-shot. Overall, this work provides a unified analysis of datasets, methods, reproducibility, and evaluation practices in medical anomaly detection, and highlights the opportunities for future research and standardized benchmarking.

Zenodo (CERN European Organization for Nuclear Research)
Anomaly Detection Techniques and Applications
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon