The He Lab is dedicated to advancing Artificial Intelligence (AI) in medical image analysis for early diagnosis / prediction of various important clinical outcomes. We are currently working on multiple National Institutes of Health (NIH)- and institution-funded studies. Our research involves diverse AI applications on medical image classification, regression, segmentation, registration, and synthesis.
Quantification of Liver Fibrosis with MRI and Deep Learning
Chronic liver disease is a common cause of morbidity and mortality in the U.S. and throughout the world. Liver fibrosis (LF) is the most important and only histologic feature known to predict outcomes from CLD. The current standard for assessing LF is the biopsy, which is costly, prone to sampling error, and invasive with poor patient acceptance. Thus, there is an urgent unmet need for noninvasive, highly accurate and precise diagnostic technologies for the detection and quantification of LF. Our overarching objective is to apply Deep Learning methods using conventional non-elastographic magnetic resonance (MR) images, MR Elastography (MRE), and clinical data to accurately detect and measure LF in children and adults with CLD, using biopsy-derived histologic data as the reference standard.
Previously, we developed a multi-channel deep transfer learning model, DeepLiverNet, to categorically classify the severity of liver stiffness using both anatomic T2-weighted MRI and clinical data for pediatric and adult patients with known or suspected pediatric chronic liver diseases. [Li, et al. 2021] This deep learning model achieved an accuracy of 88.1% in a cross-validation experiment, and outperformed our previously developed Support Vector Machine classifier. (He, et al. 2019) Such a validated model is able to triage the need for additional MREMR elastography testing, and thus potentially avoid MMR elastography in up to two-thirds of candidate patients, shortening examination length, and lowering healthcare costs.
Currently, we are developing and validating a deep learning framework to accurately segment the liver and spleen in order to extract radiomic and deep features from conventional multiparametric MRI. These features allow the detection of liver and spleen structural abnormalities/tissue aberrations. We are also developing an “ensemble” model to predict biopsy-derived LF stage using the integration of conventional multimodal MRI radiomic and deep features, MRE data, as well as clinical data.
The proposed models will help physicians to more accurately detect and follow CLD. The techniques we develop are expected to improve medical diagnosis and prognostication in the same way as deep learning has revolutionized other fields. This study will significantly impact public health because it will allow physicians and researchers to more accurately diagnose and quantify CLD and LF as well as permit more frequent assessments in a noninvasive, patient-centric manner, thus potentially improving patient outcomes while lowering healthcare costs. The techniques we develop also can be readily extended for the prediction of other important liver-related clinical outcomes, including impending complications such as portal hypertension, time to liver transplant/transplant listing, and mortality risk, among others.
NIH, R01 EB030582
MRI and Deep Learning for Early Prediction of Neurodevelopmental Deficits in Very Preterm Infants
About 100,000 very preterm infants are born every year in the United States. Up to 35% develop noteworthy neurodevelopmental deficits, thereby increasing their risk for poor educational, health, and social outcomes. Unfortunately, neurodevelopmental deficits cannot currently be reliably diagnosed until 3 to 5 years of age. Application of Deep Learning to infant brain MRI data can open up new windows into early prediction of neurodevelopmental outcomes in at-risk infants and facilitate the move towards precision medicine. Our objective is to apply deep learning to MRI acquired at term equivalent age for early prediction of neurodevelopment deficits (cognitive, language, and motor) at age 2 in VPI, and our group has identified three key components necessary for accurate prognostic models of later neurodevelopment.
By integrating multimodal MRI and clinical data, our group proposed novel end-to-end deep multimodal models to predict neurodevelopmental (i.e., cognitive, language, and motor) deficits independently at 2 years of corrected age. [He, et al, 2021]
Utilizing the local neighboring spatial information of individual voxels, we proposed a sequence of automated and objective Diffuse white matter abnormality (DWMA) segmentation approaches, including the advanced deep convolutional neural networks and 3D Residual U-Net models, to identify DWMA regions on T2-weighted MRI images for very preterm infants. [Li, et al, 2021]
More recently, we proposed a novel Ontology-guided Attribute Partitioning (OAP) method to better draw feature subsets by considering the domain-specific relationship among features. With the better-partitioned feature subsets, we developed an OAP-Ensemble Learning (OAP-EL) framework. This work was recently published in the top neuroimage journal. [Li, et al, 2022]
The techniques we develop are expected to improve the modelling fidelity in medical diagnosis/ prognosis in the same way as DL has revolutionized other fields. The DL models we develop will not only benefit early detection of neurodevelopmental deficits in VPI, but also likely benefit individuals with other neurodevelopmental and neurological diseases. This study will significantly impact public health because it will allow clinicians to target clinical and experimental intervention therapies to the most at-risk infants during periods of optimal neuroplasticity, and thus ultimately improve medical outcomes and patient well-being.
NIH, R01 EB029944
NIH, R01 NS094200
NIH, R01 NS096037
NIH, R21 HD094085