Adult primary brain tumors such as gliomas are characterized by enormous cellular diversity captured by grading. For gliomas, tumor grade is based on the region with the highest level of aberrant histopathology. Recently, a large number of subtypes have been characterized based on morphological variants and their molecular characterization showing an enormous heterogeneity that is only being discovered. This poses an enormous challenge in interpreting these subtypes and understanding their clinical and molecular associations. We propose on capturing cellular profiles in a systematic way using computational analysis of whole slide images in terms of their composition and cellular content. Current methods do not address intrinsic batch effects, biological heterogeneity that is present in a large cohort, classification of aberrant cellular morphologies (e.g., astrocytoma, oligodendroglioma), and the need to for high throughput processing of vast amount of image-based data. We propose to advance the field by addressing these issues and building a knowledge repository of brain tumor. Additionally, due to the availability of large-scale molecular data, we will build molecular signatures for prognostic morphometric features and subtypes. In this manner computational image-based modeling and representation of tumor histology can provide new avenues for hypothesis generation through molecular association. The end result will be an atlas where molecular correlates of prognostic morphometric subtypes and brain tumors can be identified. Morphometric subtypes will be validated through an independent cohort, and molecular predictors of morphometric subtypes will validated through immunohistochemistry from the sample bank at the Stanford University Medical School and University of California.
This project has major relevance for human health because it will provide tools to enable researchers to analyze biomedical image data in combination with clinical and genomics data in the TCGA resource, accelerating discovery and more rapidly leading to new strategies for treating cancer.
| Chang, Hang; Han, Ju; Zhong, Cheng et al. (2018) Unsupervised Transfer Learning via Multi-Scale Convolutional Sparse Coding for Biomedical Applications. IEEE Trans Pattern Anal Mach Intell 40:1182-1194 |
| Zhong, Cheng; Han, Ju; Borowsky, Alexander et al. (2017) When machine vision meets histology: A comparative evaluation of model architecture for classification of histology sections. Med Image Anal 35:530-543 |
| Han, Ju; Wang, Yunfu; Cai, Weidong et al. (2016) Integrative Analysis of Cellular Morphometric Context Reveals Clinically Relevant Signatures in Lower Grade Glioma. Med Image Comput Comput Assist Interv 9900:72-80 |
| Han, Ju; Fontenay, Gerald V; Wang, Yunfu et al. (2016) PHENOTYPIC CHARACTERIZATION OF BREAST INVASIVE CARCINOMA VIA TRANSFERABLE TISSUE MORPHOMETRIC PATTERNS LEARNED FROM GLIOBLASTOMA MULTIFORME. Proc IEEE Int Symp Biomed Imaging 2016:1025-1028 |
