New MRI technique May Aid Assessment of Treatment Response in Cancer Patients with Brain Metastasis

Jennifer Yu, M.D., Ph.D., and Pallavi Tiwari, Ph.D.

Cleveland Clinic Foundation

Grant Program:

David Mahoney Neuroimaging Program

Funded in:

September 2017, for 3 years

Funding Amount:


Lay Summary

New MRI technique may aid assessment of treatment response in cancer patients with brain metastasis

Investigators have developed and will validate a new method for analyzing MRI scans of cancer metastasis in the brain to see whether the metastasis are recurring following precision radiation therapy. About one of every five patients with cancers that metastasize to the brain develop some damage to brain tissues following radiation therapy. On MRI imaging, this damage is hard to differentiate from a recurrence of the metastatic cancer. Doctors then are faced with deciding whether to repeat radiation therapy based on their analyses of the imaging (which is estimated to be correct about 60 percent of the time), or perform a surgical biopsy in their weakened patients to confirm whether another round of radiation therapy is required. Neither option is optimal.

These investigators recently pioneered a new image analytics technique (called “CoLlAGe”) that computerizes subtle differences in features extracted from MRI images. This technique can show whether the tissue is disordered or organized. Disordered tissue is a biomarker for metastasis while organized tissue is a biomarker for “benign” radiation-induced tissue damage. Their preliminary data indicates that this technique accurately differentiates metastasis from radiation damage 85 percent of the time. They hypothesize that the predictive accuracy of this technique can be further increased by integrating biomarker information from three other types of MRI imaging, called T2w, FLAIR, and perfusion imaging.

To test this hypothesis, they first will create a clinically translatable “risk scoring system” based on analyses of MRI imaging and clinical outcomes from 200 prior patients. Then they will validate this risk scoring system in 200 new patients, integrating biomarker measurements obtained from the three additional forms of MRI imaging. They anticipate that a low risk score will correlate with a patient’s outcome of tissue damage from radiation therapy, while a high risk score will correlate with a patient’s outcome of metastasis recurrence.

Significance: If this combined MRI imaging and risk scoring system is validated, it would become an essential tool for safely determining if a patient has a benign radiation induced change, which could be treated with steroids, or whether metastasis have recurred and require additional treatment; thereby avoiding unnecessary brain biopsies.


New MRI technique may aid assessment of treatment response in cancer patients with brain metastasis

Background/Motivation: Brain metastases comprise about half of brain tumors and occur in about 1/3 of all cancer patients. The majority of patients are treated with stereotactic radiosurgery (SRS) as it offers high rates of tumor control by using high doses of radiation to ablate the tumor. The most challenging clinical problem in the management of metastatic brain tumors is distinguishing radiation necrosis, a side effect of SRS that affects about 20% of patients, from tumor recurrence. Patients typically receive frequent MRI scans (T1w, T2w, FLAIR and perfusion) to aid in diagnosis, with regression of the lesion and reduced perfusion favoring radiation necrosis, and enlargement of the lesion with increased perfusion favoring treatment failure. Unfortunately, the diagnostic accuracy via visual inspection by radiologists on these routine clinical MRI protocols is between 50-60%, leading to unnecessary surgical interventions in as many as 27,000 patients annually in the U.S. Advanced imaging modalities (i.e. MRS, PET) are limited by acquisition variability, cost, lack of reproducibility from site to site, and unavailability at most clinical sites. Invasive biopsy provides the only means for definite diagnosis. Therefore, finding effective, non-invasive, and economical means to differentiate between radiation necrosis and tumor progression are urgently needed. Innovation: Our team has pioneered a new radiomic technique (computerized feature extraction on radiologic images), Co-occurrence of Local Anisotropic Gradient Orientations (CoLlAGe) that can distinguish radiation necrosis from tumor recurrence using routinely acquired MRI scans. CoLlAGe captures tumor heterogeneity and lesion disorder by computing the entropy (measure of disorder) in pixel level edge directions within the lesion. Cancer is more disordered due to the breakdown of underlying tissue micro-architecture, and hence has a higher disorder in pixel orientations as compared to radiation necrosis which has a more organized micro-architecture, and hence low disorder in pixel orientations. In a pilot study of 58 lesions, CoLlAGe was able to discriminate cancer versus radiation necrosis (100% of all cancers identified) on structural MRI alone; with almost twice the accuracy of 2 expert readers. These results are very promising since none of the existing imaging modalities (including PET, MRS) have so far reported a 100% detection accuracy of tumor recurrence, with > 25% improvement in overall accuracy in distinguishing the two conditions. Research objectives: Our ultimate objective is to deploy CoLlAGe in a clinical environment as a “second read” decision support solution to assist neuro-radiologists in reliably distinguishing radiation necrosis from tumor recurrence, above and beyond what is currently possible with existing imaging protocols. Towards this objective, we first seek to perform a prospective validation study to establish the efficacy of CoLlAGe by comparing its performance with 2 expert radiologists. Additionally, we believe that combining CoLlAGe features from across different routine MRI protocols will further improve distinction of radiation necrosis from tumor recurrence. Consequently, the objective of this Dana foundation proposal is to extend our promising preliminary work by integrating CoLlAGe attributes to capture complementary entropy measurements of the tumor and peri-tumoral regions as shown on T1w, T2w, FLAIR and perfusion scans. For the ease in clinical usability, the integrated CoLlAGe measurements for every study will be projected to a scalar radiomic risk score (RRiSc) between 0 and 1, with low RRiSc signifying radiation necrosis, while high RRiSc signifying tumor recurrence. We hypothesize that the RRiSc obtained from measurements from the lesion and peri-lesion regions across T1w, T2w, FLAIR and perfusion scans, will be able to improve differentiation of radiation necrosis from tumor recurrence, as compared to (a) CoLlAGe obtained from T1w MRI alone (currently at 85% accuracy), and (b) 2 expert radiologists who independently read these scans for clinical assessment. We will evaluate the RRiSc on a curated prospective registry of SRS-treated brain metastases studies with multi-disciplinary tumor board consensus of tumor progression or radiation necrosis. Criteria for success: We have demonstrated that the accuracy of visual inspection on routine MRI scans is ~50%. With the proposed improvements, we believe that RRiSc can substantially improve the diagnosis of tumor recurrence from radiation necrosis over clinical reads in n=200 prospective studies. Specifically, our success criteria will be: RRiSc will (a) reliably classify at least 50% of those patients with radiation necrosis, and (b) have an error rate of <5% (which is lower than current clinical gold standard) meaning it will classify <5% of patients with tumor recurrence as radiation necrosis. Relevance to Dana Foundation neuroimaging grant: We seek to address the most common and challenging problem in treatment management of malignant metastatic brain tumors, using a highly innovative and novel non-invasive neuroimaging approach (e.g. CoLlAGe). In this era of value-based medicine, we propose the use of image-based risk assessment zotas high-value care for patients with metastatic brain tumors. Because only routine MRI is used, our work is highly translatable to non-tertiary care hospitals and increases the clinical and economic value of standard MRI scans (T1w, T2w, FLAIR, perfusion). With successful application of RRiSc, we do not anticipate the need for surgical interventions for disease confirmation. Patients (and society) benefit with reduced costs for a definitive diagnosis. The successful application of RRiSc could allow for near term transformative global impact by obviating unnecessary interventions and improving the value of routine MR imaging for brain tumors.

Investigator Biographies

Jennifer Yu, M.D., Ph.D., and Pallavi Tiwari, Ph.D.

Dr. Jennifer Yu, MD, PhD, is a physician-scientist at the Cleveland Clinic with dual appointments in the Department of Radiation Oncology and Department of Stem Cell Biology and Regenerative Medicine. Dr. Yu obtained her MD and PhD degrees at Columbia University, and completed her residency training in radiation oncology at the University of California, San Francisco. The mission of Dr. Yu’s laboratory is to improve therapy for patients with brain tumors by elucidating the molecular mechanisms driving cancer initiation and progression, and in doing so, promote rigorous science and train the next generation of scientists. A major focus of the laboratory is to define mechanisms underlying the tumorigenic properties of cancer stem cells with a long term goal of uncovering potential therapeutic targets. Her laboratory aims to expedite the translation of experimental therapeutics to clinical trials and improve treatment for patients.

Dr. Tiwari received her MS and PhD degrees in Biomedical Engineering from Rutgers University, after which she moved to join Case Western Reserve University as a faculty in Biomedical Engineering. Dr. Pallavi Tiwari moved to the rank of assistant professor in 2016 and since then directs the Brain Image Computing (BrIC) laboratory at Case Western. She is also an associate member of Case Comprehensive Cancer Center. Her research interests lie in pattern recognition, data mining, and image analysis for automated computerized diagnostic, prognostic and treatment evaluation solutions using radiologic imaging. Over the last 10 years, she has developed novel image analysis, and radiomics tools for diagnosis, prognosis, and treatment response evaluation of different types of cancers (prostate, breast, lung) and neuro-imaging applications including brain tumors, epilepsy, and cancer pain. She has been a recipient of Case-Coulter Translational awards from 2013 to 2015, and was twice nominated for NIH Director’s Early Independence Award. In 2015, She was named by Government of India as one of 100 women achievers for making a positive impact in the field of Science and Innovation. She is currently leading a team of researchers on evaluating prognosis, prediction, and short and long-term effects of radiotherapy and laser interstitial thermal therapy in brain tumor patients via multi-parametric MRI. Her ultimate objective is to develop clinically actionable decision support tools that could positively impact treatment management, and improve quality of life in brain tumor patients.