Joao Victor Dias

Joao Victor Dias is a statistician working at the intersection of medicine and applied machine learning, focused on medical and ophthalmic imaging. He builds deep-learning pipelines for image reconstruction and segmentation, including cross-vendor retinal OCT segmentation for a MICCAI 2026 challenge track, where the core problem is recovering a faint, clinically meaningful signal from noisy, blurred, multi-channel data acquired on heterogeneous devices, often with little or no ground truth. That is exactly the problem hyperspectral biology faces at a distance: detecting a weak engineered reporter against an unknown natural background, in low-SNR cubes from cheap or far-away sensors, with only a handful of measured reference spectra. He has shipped reproducible models on free and low-cost GPUs (Kaggle, Colab), which keeps this project affordable and easy for others to reuse. This is an individual, non-commercial project: all code, the scene simulator, the spectral dataset and the benchmark will be released under the MIT license and archived with a DOI, so any researcher, university lab, community lab, or dedicated amateur, can build on them.

June 2026