Sina Ghandian

Spearheaded a multi-institution collaboration aiming to segment neurofibrillary tangles (NFTs) in gigapixel whole slide images by creating a novel and scalable pipeline to convert point annotations into ground truth masks, enabling efficient active learning. ● Delivered a platform presentation at the American Assoc. of Neuropathologists 99th Annual Meeting detailing our trained segmentation model’s performance (AUC 0.83) and its strong alignment (rho=0.654) with pathologist semi-quantitative grading. ● Developed an efficient method for generating high-resolution prediction heatmaps via PyTorch-Lightning and Zarr file-locking; applied it to cerebral amyloid angiopathy detection; shared findings at the 2022 International Conference for Systems Biology. ● Collaborating to train a pair of neural networks to detect melanoma in dermatopathology images stained with two different immunohistochemicals; applied saliency mapping to create an in-silico stain of melanoma from a tile-level classification model. ● Autonomously scoping out project directions and developing codebases & version-controlled documentation via GitHub. ● Mentoring and supporting colleagues in their projects, fostering their growth, and leading projects toward completion.