Deep learning for medical image analysis (computational pathology)
Digital pathology images are gigapixel scans, far too large to feed to a model in one piece, so the work starts by finding the tissue and cutting it into patches. On those patches I train and compare CNNs (ResNet, U-Net, DeepLabV3+) and Vision Transformers (SegFormer and pathology foundation models used as frozen encoders with light decoders, plus LoRA fine-tuning) for semantic segmentation and object detection. Everything runs in PyTorch as reproducible, config-driven experiments tracked in MLflow, on a GPU workstation and an HPC cluster. The research is not published yet, so this page describes the field and the techniques only.
Details after publication.
