Imaging Analysis for Precision Oncology
We are a leading group in radiomics research, which is aimed to improve prognostication and prediction of treatment response for personalized cancer therapy. Our approach incorporates deep learning as well as knowledge-based profiling of the intratumoral subregions/habitats and surrounding parenchyma. We have conducted imaging biomarker studies on breast, lung, head and neck, and gastric cancers, using several radiologic modalities including PET, CT, and MRI (Cui et al. Radiology, 2016; Wu et al. Radiology, 2016, 2018; Wu et al. J Nucl Med, 2019, Jiang et al. Annals Surgery, 2020, Jin et al. Nature Communications, 2021). In a recent study published in Nature Machine Intelligence, we developed a unifying radiological tumor classification system that demonstrates prognostic relevance and therapeutic implications across multiple cancer types.
A major focus of our lab is artificial intelligence for histopathology image analysis. In ongoing work, we are developing the next-generation machine learning and deep learning approaches to analyze and extract information from histopathologic images including routine H&E-stained slide and multiplexed immunofluorescence imaging. We are working with a team of expert clinicians and pathologists toward the goal of predicting immunotherapy response and patient outcomes.
Integrative Radiomics and Immunogenomics
We are also exploring the biological basis of clinically relevant radiologic phenotypes by correlating imaging with molecular data (Wu et al. Radiology, 2017; Wu et al. Clin Cancer Res, 2017).
In recent studies, we developed a noninvasive radiographic approach to evaluate tumor immune and stromal microenvironment, and showed these signatures could predict chemotherapy response and survival outcomes (Jiang et al. Annals Oncology, 2020; Jiang et al. Lancet Digital Health, 2021).
In addition to imaging, we are also developing molecular biomarkers to predict therapy response and prognosis in cancer. We are particularly interested in how various components of the tumor microenvironment (including immune and stroma) interact with each other and contribute to cancer progression or control.
Previously we developed an individualized immune gene expression signature that predicts survival of patients with non-small cell lung cancer (Li et al. JAMA Oncology, 2017). In another work, we integrated radiation sensitivity and immune signatures to predict which patients are most likely to derive survival benefit from adjuvant radiotherapy in breast cancer (Cui et al. Clin Cancer Res, 2018). In a study of pan-squamous cell carcinomas (including esophagus, head and neck, lung, and cervix), we identified 6 immune subtypes associated with distinct molecular characteristics and clinical outcomes (Li et al. Clin Cancer Res, 2019).
In ongoing work, we are investigating machine learning approaches for analyzing genetic and epigenetic profiles and developing molecular biomarkers for early cancer detection, prognostication, and response prediction.