Group members: Brandon Fischer and Ryan Danehy
Our proposed project, in partnership with Dr. Don Miller's lab from the IU School of Optometry, will be a binary classifier which is trained to differentiate (and generate a count of) S-cones from L and M cones in 3D retinal imaging. We plan to employ an RBF-kernel SVM to classify S-cones vs non-S-cones. This project has clinical significance in the tracking of progression of the disease Retinitis Pigmentosa (RP). In RP, S-cones can be seen migrating from their natural positions, and eventually disappearing entirely in retinal scans. Our service could be extended from purely classifying/counting S-cones to tracking the rate of their movement and determining the progression/severity of the disease in a given patient.
Datasets will be provided by Dr. Miller's lab which will include the 3D coordinates and aperture size of each cone detected within the retinal scan. Using this information, we should be able to differentiate the S-cones from the others due to their deeper position and wider aperture compared to the other cell types. Our starting dataset includes information from the images of 3 patients' retinas, and will be divided into sub-regions to give us more subsets. Additional data may be collected/requested as needed.
After the initial model is trained, unlabeled data can be given for classification via rest service, and the count of S-cones (and any other pertinent information) will be returned via another rest service.