Despite advances in therapy, the most aggressive form of brain tumor, glioblastoma, remains almost universally fatal. The first-line therapy for this devastating cancer is maximum feasible surgical resection, followed by radiotherapy with concurrent temozolomide chemotherapy (CRT). It is encouraging that there are multiple second-line therapies in clinical trials that could improve life quality or prolong survival, such as anti- angiogenic therapy (AAT). In this scenario, the accurate determination of whether a patient is a responder or a non-responder at an early stage following CRT has become a significant factor in clinical practice. However, the limitations in neuroimaging complicate the clinical management of patients and impede efficient testing of new therapeutics. Even with the improvements in advanced imaging modalities, distinguishing true progression vs. pseudoprogression (induced by CRT), or response vs. pseudoresponse (induced by AAT) remain two of the most formidable diagnostic dilemmas. Hence, the current gold standard for diagnosis and local therapy planning is still based on pathologic appraisal of tissue samples. However, even this yields variable results due to the intra-tumoral heterogeneity of treatment response. Therefore, reliable imaging tools, capable of early prediction of the tumor response to clinical therapies, are urgently needed. Amide proton transfer-weighted (APTw) imaging is a chemical exchange saturation transfer (CEST)-based molecular MRI technique, which has been demonstrated to add important value to the clinical MRI assessment in neuro-oncology. However, most currently used imaging protocols are essentially semi-quantitative, and the images obtained are often called APTw images because of other contributions. Notably, it has been shown that quantitative CEST-MRI is able to achieve more pure and higher APT signals in patients with brain tumors. On the other hand, deep- learning is a state-of-the-art imaging analysis technique that provides exciting solutions with minimum human input. In particular, the saliency maps derived act as a localizer for class-discriminative regions, and may have great potential to guide biopsies and local treatment regimens. The goals of this proposal are to demonstrate the potential of quantitative CEST-MRI to resolve two formidable diagnostic dilemmas for GBM patients and to develop an automated deep-learning framework for post-treatment surveillance and biopsy guidance. This application has three specific aims: (1) Implement and optimize the quantitative CEST-MRI technique and quantify its accuracy in predicting early response to CRT and survival; (2) Determine the capability of quantitative CEST-MRI to assess the response to bevacizumab; and (3) Develop a deep-learning pipeline that includes structural and CEST images for responsiveness differentiation and stereotactic biopsy guidance. If successful, our results?and particularly the deep-learning platform established?will be readily available to accurately identify early response and guide stereotactic biopsy, thus changing the clinical pathway.
Distinguishing recurrent tumor from treatment effect following therapy remains a major clinical challenge in neuro-oncology. We will assess the potential of quantitative CEST imaging methodologies and develop an automated, deep-learning framework for post-treatment surveillance and biopsy guidance. If successful, the implications for the clinical management of patients with brain tumors and for the robust evaluation of the efficacy of experimental therapeutics are enormous.
