Michael Li
Medical student
About
Michael Li is a medical student who writes software. The projects below were checked against data they had not seen, or run for real, and each write-up ends with what the thing still gets wrong. Two are imaging classifiers for the clinic: one tells apart the contrast phases of an abdominal CT from two vessel intensities, one scores dermoscopic lesions on an outside image set. Two run on one laptop for one person: a lecture-to-flashcard tool that holds back any card it cannot trace to a spoken sentence, and an assistant that reads calendar, mail and saved posts and cannot write to any of them.
Chosen works
| Year | Title | Summary | Domain | Status |
|---|---|---|---|---|
| 2025 | Melanoma-Li | Dermoscopic lesion classification, tuned by nested cross-validation and scored on 1,512 outside images. | Deep learning, PyTorch | Case study |
| 2025 | AccessKit | A PDF reader and a music controller driven by ten facial gestures and a webcam. | Computer vision, Python | Case study |
| 2024 | PRISM | CT contrast-phase classification from two vessel intensities, checked on 827 scans from other institutions. | Medical imaging, MATLAB | Case study |