AI-Driven Medical Imaging System for Health Disparity Detection
SUMMARY
Medical imaging artificial intelligence probe offering enhanced detection, continuous monitoring, and quantitative analysis of health disparities across populations by leveraging large datasets and advanced machine learning to extract radiomic features from routine medical images, thereby supporting more precise identification of inequities embedded in imaging data
The Unmet Need: Limited tools exist for scalable, objective detection of health disparities from medical images
- Health disparities, which represent differences in disease burden among population groups often linked to social determinants of health (SDOH), remain inadequately characterized due to the limitations of existing data sources such as electronic health records and manually generated clinical reports. These sources often lack comprehensive socio-demographic information and are not available at the pixel or image phenotype level, constraining researchers’ ability to perform large-scale and granular disparity analyses. Consequently, there is a significant gap in objective, automated approaches that can detect and quantify health disparities embedded within routinely acquired medical imaging data.
-
Advances in artificial intelligence and machine learning applied to medical imaging, coupled with increasing availability of large, standardized imaging datasets in DICOM format, have created an opportunity to integrate computational radiomic analyses into health disparities research. The trend towards population-level imaging data analytics and automated feature extraction supports the development of scalable, continuous monitoring systems that extend beyond traditional clinical and demographic data paradigms, facilitating more timely and geographically detailed disparity surveillance.
The Proposed Solution: Development of an AI/ML-based medical imaging probe utilizing radiomic phenotypes to quantify health disparities at the population level
- The faculty inventor developed an artificial intelligence system which applies established deep learning and radiomic feature extraction methodologies to large medical image datasets for the purpose of detecting and monitoring health disparities. The core mechanism involves the automated derivation of quantitative imaging phenotypes- computer-derived radiomic features- from routine medical images such as chest radiographs. These features are subsequently analyzed using statistical and machine learning techniques, including principal component analysis, Hotelling’s t-squared statistics, and clustering algorithms, to stratify populations into radiomic phenogroups and generate imaging-derived disparity indices aligned with frameworks like the National Cancer Institute’s HD*Calc.
- Unlike existing image-based diagnostic tools focused on disease detection or risk assessment, this technology uniquely targets disparity evidence embedded in imaging data for continuous population-level surveillance. Preliminary validation leveraged de-identified institutional chest radiographs to demonstrate feasibility in deriving novel health disparity metrics.
FIGURE
Conceptual overview of the medical imaging ‘big data’ health disparities (HD) platform for (a) the repurposing of ‘normal’ images for continuous monitoring of HDs and (b) an additional data source for HD research
ADVANTAGES
-
Automated extraction of imaging-based health disparity metrics
-
Scalable analysis of large medical image datasets
-
Objective quantification of radiomic phenotypes
-
Continuous population-level disparity monitoring
-
Integration with existing disparity metric frameworks
-
Non-invasive use of routine clinical imaging data
APPLICATIONS
-
Health disparities research in lung disease using chest radiographs
-
Population-level monitoring of demographic inequities in medical imaging outcomes
-
Supplementing socio-demographic data for epidemiologic and clinical disparity studies