Quantitative Imaging, Radiomics, and Machine Learning
We apply quantitative imaging, radiomics, and modern machine learning techniques to study several cardiothoracic imaging: including lung cancer, interstitial lung disease (including interstitial lung abnormality), and emphysema. Additionally, we have completed a randomized controlled trial on 'rapid rollover' technique to reduce pneumothorax following lung biopsy. We have multiple collaborative computer vision and clinical research projects on screening, diagnostic, and intervention around lung cancer imaging.
Radiological Natural Language Processing
We are part of the UCSF Center for Intelligent Imaging. For Natural Language Processing, we study radiological reporting, patient/provider communication, and text based data in radiology leveraging modern natural language processing techniques including but not limited to developing and translating large language model, large vision language model, and radiological report optimization. This is in strong collaboration with UC Berkeley and industry scientists.
Mid-field Lung and Cardiac MRI
We conduct clinical and translational research on novel imaging modalities including the 0.55T Lung MRI and other emerging modalities in collaboration with imaging scientists and industry collaborators. We have scanned nearly 100 patients on the novel 0.55T lung MRI system with protocols and techniques optimized in-house in collaboration with imaging scientists. We have also developed a rapid combined cardiac, pulmonary, and ventilation/perfusion imaging for various pulmonary pathologies.

Cardiothoracic Imaging and Natural Language Processing Laboratory

Welcome to the Sohn lab! 

Our lab specializes in using modern data science and advanced computing techniques to improve the practice of cardiothoracic imaging.

We study various imaging modalities, including radiographs, CTs, and MRIs of the heart and lungs. We explore all clinical topics in cardiothoracic imaging, but we are especially interested in lung nodules/cancers, emphysema, and interstitial lung disease. Furthermore, our lab's additional focus is on improving ways in which radiologists communicate with referring clinicians and patients. We have conducted both technical and clinical validation work on radiological reporting with the assistance of modern natural language processing techniques such as large language models. 

We are, by design, a highly collaborative and multidisciplinary team, very well connected with other radiologists, clinicians, imaging scientists, and machine learning engineers around the country and the world. Thank you for visiting our website and please reach out with any questions!

Best,
Jae Ho Sohn, MD, MS