selected projects
Kaggle competition solutions, medical imaging research, and deep learning tools.
Browse my GitHub repositories and Hugging Face models and demos.
Kaggle competitions
Source code from my competition work, including team solutions hosted by collaborators.
Pneumonia detection · 1st place
Detecting and localizing pneumonia on chest radiographs for the RSNA Pneumonia Detection Challenge.
Pulmonary embolism detection · 2nd place
Detecting pulmonary emboli on CT using image models and sequence modeling for the RSNA STR competition.
Melanoma classification · 2nd place
Classifying skin lesions for the SIIM-ISIC Melanoma Classification competition, using external training data and pseudo-labeling.
Breast cancer detection · 4th place
Our team’s solution for detecting breast cancer on screening mammograms in the RSNA competition.
Cervical spine fracture detection · 6th place
A pipeline combining vertebral segmentation, 3D classification, and sequence models to detect cervical spine fractures on CT.
Pneumothorax segmentation · 8th place
Identifying and segmenting pneumothoraces on chest radiographs for the SIIM-ACR competition.
Intracranial aneurysm detection · 6th place
Detecting intracranial aneurysms for the RSNA Intracranial Aneurysm Detection competition.
Abdominal trauma detection · 8th place
Detecting abdominal injuries on CT for the RSNA 2023 Abdominal Trauma Detection competition.
COVID-19 detection · 3rd place
Classification and detection models for the SIIM-FISABIO-RSNA COVID-19 competition. The team repository includes my training pipeline alongside my collaborator’s work.
Lumbar spine · 2nd place
Training code and experiment configurations for the RSNA lumbar spine competition.
Research and tools
SKPv2 · Deep learning experimentation
My framework for rapidly prototyping deep learning experiments, with flexible Python configurations and a project template designed for Kaggle-style iteration.
Pediatric bone age assessment
Hand localization and bone age estimation from radiographs, with training code, inference code, and pretrained models.
Code · Hugging Face model · Demo
Active learning for diabetic retinopathy
Experiments studying how to select retinal photographs for labeling efficiently, accompanying our paper on deep active learning.
Chest X-ray lines and tubes
A RANZCR-based project combining catheter and tube classification with segmentation of lines, endpoints, and the carina.
Code · Hugging Face model · Chest X-ray demo
Head CT hemorrhage detection
Research models for detecting intracranial hemorrhage on head CT, with predictions at the slice and scan level and visualizations of hemorrhage location.
Code · Hugging Face model · Demo
Screening mammography
Models and a demo for breast cancer risk and breast density prediction from mammograms.