04.1
Hyperspectral tissue
My partThe classifiers, trained and tuned in Python.
Hyperspectral imaging is researched as a way to find cancer during surgery: a tumour can look like healthy tissue and still differ in its spectrum. A hyperspectral camera records each pixel as a spectrum of 92 narrow bands. I trained models to tell healthy tissue from cancer by that spectrum alone, from classical classifiers to neural networks. The best, a random forest, reached an F1 score of 0.934.
λ 400 nm
Drag along the spectrum to read one band. The tissue is drawn, not a patient's.
