Harrison Bai, MD
Assistant Professor of Diagnostic Imaging
Clinician Educator
Brown University 

Congratulations to Harrison Bai, MD, who has been awarded a Research Scholar Grant from the Radiological Society of North America (RSNA). 

"Deep Learning-based Response Assessment and Outcome Prediction for Transarterial Chemoembolization Treatment of Hepatocellular Carcinoma"

Dr. Bai submitted his RSNA application as part of the 2019 Advance-K Scholar Career Development Program. The yearlong program from the Brown Division of BioMed and Advance-CTR provides one-on-one and group training to junior investigators with the goal of submitting a successful NIH K or equivalent proposal by the conclusion of the program. 

The Study 

Hepatocellular carcinoma (HCC) is one of the leading causes of cancer-related burden worldwide. Current therapies of HCC include localized resection, transplantation and transarterial chemoembolization (TACE). However, due to the rapid growth of HCC tumors, most patients are diagnosed with intermediate stage disease at presentation, for which TACE is the current standard therapy. Imaging, in particular arterial-phase enhancement on MR image, plays an important role in assessing the therapeutic response of treatment. Enhancement-based response criteria, such as modified RECIST (mRECIST) and The European Association for the Study of the Liver (EASL), have been shown to predict the long-term survival for HCC patients undergoing TACE.

However, the current response criteria, assessed by human observers (2D) or software (3D), sometimes fail in inter-reader reproducibility, accuracy in detecting treatment response, and, most importantly, prediction of a survival benefit. This lack of precision and consistency in response assessment makes both prognostication and treatment planning difficult for patients who have undergone TACE. Furthermore, these criteria are based on post-treatment imaging appearance. The inability to accurately determine which patients will benefit the most from TACE based on characteristics on pre-treatment imaging and/or clinical information makes it challenging to appropriately select patients for therapy and strategize clinical-decision making.

To address these issues, we propose to develop novel deep neural network models that allow for accurate (1) assessment of treatment response that is superior to the current standard and (2) prediction of treatment response and survival based on routine pre-treatment MRI and clinical variables. Our hypothesis is that deep learning assessment of treatment response will correlate with patient survival better than the current 2D and 3D standards. In addition, deep learning based on pre-treatment MRI will allow for highly accurate response assessment and outcome prediction that could reduce unnecessary interventions, lower health care costs, and minimize patient harm.  

Keys to Success

"Advance-K helped me significantly as the faculty leads connected me with the correct mentor, and gave me feedback on various aspects of my project. 

I developed relationships that significantly helped with my subsequent research. For example, the mentors I got to know introduced by the Advance-K faculty developed an interest in my career and started to advise on applying to other grants and presenting at national conferences. Advance-CTR connected me with a statistician who has since worked with me on multiple grants and papers. I started to collaborate with several fellow junior faculties in the Advance-K program.

I am extremely grateful for how much Advance-K and Advance-CTR helped me in my career development."