This technology is a statistical harmonization method that addresses batch differences in radiomic images, such as CT datasets, taken at different institutions, allowing for comparison and generalization.
Radiomic analysis is a technique that analyzes large amounts of quantitative data from medical images, such as CT scans, capturing subtle shapes, textures, and intensities that can be linked to clinical outcomes. Therefore, it has significant potential to advance precision medicine and other clinical applications; however, its widespread use is currently hindered by variability in image acquisition across scanners and institutions, which introduces non-biological differences. Current harmonization methods, such as ComBat, provide partial correction for these batch effects but are limited by the assumptions of normally distributed data, dependence on known imaging parameters, and the inability to process multimodal sources of variation. Overcoming these limitations is crucial for enhancing the robustness and generalizability of radiomic biomarkers across diverse imaging environments and study designs.
This technology improves existing radiomic harmonization techniques by introducing optimized, iterative methods that can correct for multiple and unknown sources of imaging variability. It builds on the ComBat framework by automatically determining the optimal sequence for harmonizing batch effects and by integrating a data-driven approach to identify and adjust for hidden covariates that cause multimodal feature distributions. Together, these improvements enable more accurate standardization of radiomic data across diverse imaging conditions while preserving clinical variation.
This technology has been validated using two publicly available computed tomography datasets of lungs.
Proxy118
Licensing Contact: Joan Martinez
