How to Transform AI Prototypes into Functional Healthcare Applications for Diagnostic Assistance?
Presented During:
Thursday, June 27, 2024: 11:30 AM - 12:45 PM
COEX
Room:
Grand Ballroom 103
Poster No:
697
Submission Type:
Abstract Submission
Authors:
Martin Dyrba1, Devesh Singh1, Doreen Goerss2, Olga Klein1, Marc-André Weber3, Stefan Teipel4
Institutions:
1German Center for Neurodegenerative Diseases (DZNE), Rostock, Germany, 2Rostock University Medical Center, Rostock, Germany, 3Institute of Diagnostic and Interventional Radiology, Pediatric Radiology and Neuroradiology, Rostock, Germany, 4Rostock University Medical Center & German Center for Neurodegenerative Diseases (DZNE), Rostock, Germany
First Author:
Co-Author(s):
Marc-André Weber
Institute of Diagnostic and Interventional Radiology, Pediatric Radiology and Neuroradiology
Rostock, Germany
Institute of Diagnostic and Interventional Radiology, Pediatric Radiology and Neuroradiology
Rostock, Germany
Stefan Teipel
Rostock University Medical Center & German Center for Neurodegenerative Diseases (DZNE)
Rostock, Germany
Rostock University Medical Center & German Center for Neurodegenerative Diseases (DZNE)
Rostock, Germany
Introduction:
As the number of elderly people is rapidly increasing, we are facing a higher demand of diagnostic services, for example to detect neurodegenerative diseases. At the same time, the number of medical centers and experts remains almost constant, which poses a challenge. Tools for diagnostic assistance are urgently needed to improve the efficiency of healthcare. In three externally funded projects, we investigate aspects and strategies of how artificial intelligence (AI) prototype systems can be translated into functional healthcare applications.
Methods:
In the ongoing project "ExplAInation" funded by the German research foundation (DFG), we have developed a deep learning framework that generates visual and textual explanations to improve the comprehensibility and interpretability of such models (Dyrba et al. 2021). A prototype system that supports the evaluation of MRI scans for the diagnosis of dementia is currently being evaluated. Within the project "Clinical AI-based Diagnostics" (CAIDX), funded by the European Interreg Baltic Sea Region program, we are investigating barriers and best practice recommendations for the integration of AI prototypes or commercial tools in hospitals. To this end, we have conducted initial interviews with various stakeholders from companies, developers, hospital administration, and clinical users. In the complementary project "TESIComp", funded by the Federal Ministry of Education and Research (BMBF), we are investigating ethical and social aspects of diagnostic AI tools. For the dementia use case, we interviewed patients, caregivers, and doctors from a memory clinic and will conduct focus groups and an observational study to examine how these tools may change the patient-physician relationship.
Results:
We developed a deep learning application for the detection of dementia atrophy patterns in brain MRI scans (Fig.1). Derived relevance maps and textual summary reports highlight diagnostically important atrophy patterns for further evaluation by the physicians. Preliminary results from interviews with clinicians showed an explicit desire for AI tools in order to reduce the workload by taking over repetitive tasks. Clinicians stated a reliable performance of results, and robust and efficient usability as prerequisites for regular use. The focus groups and observational study will elucidate future changes and challenges in the doctor's role and responsibilities and highlight ethical and social considerations with respect to the use of AI tools in clinical practice.
Conclusions:
In "Healthcare 3.0", the digital transformation will change current diagnostic procedures and roles. Our activities focus on the stakeholders involved as well as on the regulatory aspects and implementation strategies to better steer this process. In CAIDX, we will develop best-practice guidelines for the overarching process of integrating AI tools in the hospital. In TESIComp, we will help to assess the future influence of AI-based approaches in clinical practice and will provide empirically informed ethical recommendations.
Education, History and Social Aspects of Brain Imaging:
Education, History and Social Aspects of Brain Imaging 1
Modeling and Analysis Methods:
Classification and Predictive Modeling 2
Keywords:
Computational Neuroscience
Data analysis
Degenerative Disease
Design and Analysis
Machine Learning
MRI
Open-Source Software
Social Interactions
1|2Indicates the priority used for review
·Fig.1 Interactive visualization app for the AI evaluation of MRI scans to highlight patterns of Alzheimer’s disease