List view
This is a collaborative effort with the QuCAD Team, led by Elim Thompson, where we aim to take the AI performance values from our trained CADt model [ICH-CADt](https://github.com/jmweaver-FDA/ICH-CADt) and evaluate it on real patient data and synthetic data to assess clinical impact in terms of time savings under different scenarios. Using the [synthetic CT imaging pipeline](https://github.com/DIDSR/PedSilicoICH) we can control different scenarios in terms of patient, disease, and imaging distributions/prevalence to see how that impacts patient time savings for this time sensitive pathology, intracranial hemorrhage. This is related to project Aim 2 on task-based validation of our synthetic data where we aim to compare estimated time savings and performance to real data and trends from the literature.
Overdue by 7 month(s)•Due by October 10, 2025•2/2 issues closedThis can include task-independent or task-dependent evaluations. Task-independent could involve comparing distributions of measures variables such as lesion contrast, volume, and shape and defining an acceptance criteria for how close the real and synthetic distributions need to be, e.g. same mean within +- 1 std. Task-dependent involves a reader, this reader could be a human reader, a radiologist, or an algorithmic observer, such as an AI detection model. This validation can include subgroup analysis by demographics (adults vs peds), lesion characteristics (large vs small lesions), or acquisition (dose level, recon type, vendor model)
Overdue by 7 month(s)•Due by October 10, 2025•4/8 issues closedConsistent with our Skynet milestone, we need to determine our critical to quality here, and what are essentials for our ICH models
Overdue by 1 year(s)•Due by December 20, 2024•7/9 issues closed