Unmet Need
As AI continues to integrate into medical imaging techniques, the ability to effectively utilize the millions of data points from those scans to effectuate future treatment innovations is critical. Radiomics is widely acknowledged for its potential impacts in inferring underlying patient conditions and accurately predicting patient outcomes, yet it is inherently hindered in a few regards. Variability in radiomics values may arise through: 1) data collection techniques from various imaging systems and their associated parameters; 2) an unaddressed standardization in how images are processed and subsequent computation of features; and finally, 3) a lack of uniformity in how radiomics models are reported. The need exists for a practical software solution that standardizes radiomics workflow, creates repeatable and impactful results across imaging modalities, and integrates into current processes with the potential to harness future AI innovations. The result would be a robust, streamlined process that supports physician decision-making and continually improves patient treatment and outcomes.
Technology Overview
Johns Hopkins researchers have created a software framework capable of predicting the mean and variability of radiomics based on an image's properties. The prediction framework is also used to recover measured radiomics to a ground truth value or the value measured from a different imaging protocol. Imaging modalities are diverse in their capabilities. Thus, this framework provides capabilities for linear imaging systems and systems where noise and resolution are calibratable, or even when there is limited knowledge of the model and the image-dependent noise/resolution. This software framework provides a predictive system that considers full system resolution and noise properties and promotes further standardization in medical imaging.
Stage of Development
Proof of concept. Over the next four years, we intend to validate this approach at three different sites across a range of CT scanner devices.
Patents
Provisional patent application 63/121,646 filed on 12/04/2020.
