Related Experiment Video
Updated: May 17, 2025
07:23
Optimization of the Retinal Vein Occlusion Mouse Model to Limit Variability
Published on: August 6, 2021
2.6K
Minimizing Human-Induced Variability in Quantitative Angiography for Robust and Explainable AI-Based Occlusion
Parmita Mondal1,2, Mohammad Mahdi Shiraz Bhurwani3, Swetadri Vasan Setlur Nagesh2,4
1Department of Biomedical Engineering, University at Buffalo, Buffalo, NY 14260.
Arxiv
|March 31, 2025
Summary
Bias in quantitative angiography (QA) can affect intracranial aneurysm (IA) occlusion prediction. Correcting this bias significantly improved deep neural network (DNN) accuracy for predicting IA occlusion after flow diverter treatment.
Area of Science:
- Medical Imaging Analysis
- Artificial Intelligence in Medicine
- Vascular Surgery
Background:
- Contrast injection variability in quantitative angiography (QA) poses a significant challenge for accurate intracranial aneurysm (IA) occlusion prediction using deep neural networks (DNNs).
- Existing methods struggle with inherent biases in imaging data, impacting the reliability of AI-driven outcome predictions.
Purpose of the Study:
- To implement an injection bias removal algorithm to reduce QA variability.
- To evaluate the impact of explainable AI (XAI) on the reliability and interpretability of DNN models for predicting IA occlusion post-flow diverter treatment.
Main Methods:
- Angiograms from 458 patients with flow diverter-treated IAs were analyzed.
- Injection variability was minimized by deconvolving and reconvolving the parent artery input with a standardized curve.
- A DNN model was trained on QA biomarkers for six-month occlusion prediction, with Local Interpretable Model-Agnostic Explanations (LIME) used for feature identification.
Main Results:
- The DNN model achieved an AUROC of 0.60±0.05 without bias correction.
- Post-correction, the DNN's AUROC increased to 0.79±0.02, with accuracy rising from 0.58±0.03 to 0.73±0.01.
- Sensitivity and specificity were 67.61±1.93% and 76.19±1.12%, respectively, with LIME enhancing model interpretability.
Conclusions:
- Standardizing QA parameters through injection bias correction enhances the accuracy of IA occlusion prediction in flow diverter-treated cases.
- Integrating XAI methods like LIME improves the transparency and clinical relevance of AI models for outcome prediction.
- This study demonstrates the feasibility of developing clinically interpretable AI solutions for predicting treatment outcomes in neurovascular interventions.
