AANS Neurosurgeon | Volume 28, Number 1, 2019


Data, Analytics and the Future of Neuro-oncology

Print Friendly, PDF & Email

The opportunities and challenges associated with “big data” have been major topics in healthcare for the past 15 years. Similarly, “Artificial Intelligence,” has recently emerged as the latest of the technology-associated buzzwords to enter the lexicon of clinicians and medical investigators. Both are, fundamentally, two sides of the same  coin. The former describes the technology and science of aggregating, organizing and making available to end users large-scale sources of comprehensive information, while the latter describes an emerging set of computational techniques capable of identifying and making decisions based on patterns in these data sets that may not be readily apparent to the analyst. As data stores and computational tools have become more available, clinicians and researchers have worked to capitalize on these resources to ask and answer questions that were previously unanswerable. 

Neuro-oncology has been no exception to this approach, and its practitioners continue to look to the age of data and computation with optimism. As we enter an era where data availability is less of a problem and data-driven question answering is gaining a more solid computational footing, it is logical to ask what may be the next major areas of growth in data-driven neuro-oncology. At least three general areas of clinical neuro-oncology are likely to see significant growth and change over the next decade because of the growth of data and computational resources:

  • Practice of diagnosis and classification of brain tumors;
  • Pre-intervention prognostication and risk stratification for patients with brain tumors; and
  • Developing individualized treatment strategies for specific patients that lead to optimal outcomes.

All are worth consideration.

Diagnosis and Classification of Brain Tumors

This area of study has traditionally been the purview of neuropathologists, but the future may hold an increased emphasis on automation and standardization of histologic diagnosis supplemented by additional diagnostic data garnered from radiologic and molecular biologic domains. The latest WHO Classification of Brain Tumors has, for the first time, made meaningful inclusion of a few molecular markers in the classification schema for gliomas1. This trend toward inclusion of non-histologic diagnostic features seems likely to continue and grow as more comprehensive molecular profiling (already commoditized as a service by Caris and Foundation 2,3) becomes a standard part of brain tumor pathologic examination. Enhanced MRI imaging capturing more biologically-relevant information will also have enormous impact. Already automated imaging classifiers4,5 and radiogenomic approaches to brain tumor classification6 are demonstrating the potential accuracy and value of these techniques. Novel classification schema based upon imaging patterns recognized by AI systems may augment existing methods of histopathologic and molecular classification. The near future holds the promise that the brain tumor diagnosis may become more timely and accurate with the inclusion of automated, AI-assisted, data-driven technologies.

Pre-intervention Prognostication and Risk Stratification

For decades information regarding the risks and expected outcomes of cancer interventions has been based on generalization of median values calculated by descriptive analytics of large patient cohorts and databases7/span>. More recently, personalized prognostic modeling has improved the process of informed decision-making regarding the role of surgery and the balance between risk and extent of tumor resection for an individual patient8,9. The next iteration of these approaches will involve integration of more comprehensive data sources and more complex models of tumor pathobiology10. Clinicians could then have individualized and more accurate discussions with patients regarding the relative risks and benefits of various interventions for management of their unique brain tumor. This, in turn, can help manage expectations and optimize patient-specific preferences with regard to the balance between efficacy and quality of life.

Individualized Treatment Strategies

The Holy Grail of personalized medicine is using data from multiple sources in conjunction with AI algorithms to develop individualized treatment strategies for specific patients that lead to optimal outcomes.  Given the degree of molecular heterogeneity of many brain tumors11 and the variability in survival and response to therapy seen among patients with a given tumor type, it is reasonable to hypothesize that a single, one-size-fits-all treatment  may not be  ideal for every patient. Domain-specific studies in AI-assisted personalized therapy have demonstrated feasibility of this approach in radiotherapy planning12 and in chemotherapy dosing 13,14 in brain tumor patients. A logical continuation of this work will be to validate these models and then to extend the methodology to treatment protocols that include combinations of therapeutic modalities. This would yield AI tools that draw on multiple sources of clinical, molecular and diagnostic data to create personalized therapeutic regimens for individual patients. This approach has long been a goal for neuro-oncologists, and recent advances in data science and artificial intelligence, in combination with novel approaches to AI-based adaptive clinical trial design15, may finally see the dream coming to fruition.

Neuro-oncology Science of Practice for Tomorrow

Diagnosis, prognostication, risk stratification and personalized therapy for brain tumors are poised to benefit from the growing applications of big data and AI in study design and clinical decision-making. It is important to note, however, that much of the current excitement regarding these applications is appropriately relegated to the categories of buzzwords and hype. While promising new work is demonstrating the art of the possible, it is more critical than ever to insist upon careful adherence to the scientific method and a healthy degree of skepticism if these new technologies are ultimately to lead to clinical management strategies that are both safe and effective. The quality of the underlying data remains paramount, and rigorous theoretical understanding of analytic methods that are progressively becoming more black box have never been more necessary. If a new incarnation of traditional scientific rigor can be applied appropriately to these emerging technologies, there is significant potential for computational analytics to drive improvements in our understanding of brain tumors and in the management of neuro-oncology patients. 

The future lies not in big data and AI in and of themselves, but rather in the intelligent and rigorous application of these technologies by careful and thoughtful clinicians and investigators. 


Click to View All

1. Louis, D.N., et al. (2016). The 2016 World Health Organization Classification of Tumors of the Central Nervous System: a summary. Acta Neuropathol. 131(6):803-20.  

2. https://www.carismolecularintelligence.com/ 

3. https://www.foundationmedicine.com/

4. https://www.xinhuanet.com/english/2018-06/30/c_137292451.htm

5. https://biomind.ai/

6. Kickingereder, P. et al. (2016). Radiogenomics of Glioblastoma: Machine Learning-based Classification of Molecular Characteristics by Using Multiparametric and Multiregional MR Imaging Features. Radiology. 281(3):907-918. 

7. https://www.cbtrus.org/factsheet/factsheet.html 

8. Marko, N.F. et al. (2014). Extent of Resection of Glioblastoma Revisited: Personalized Survival Modeling Facilitates More Accurate Survival Prediction and Supports a Maximum-Safe-Resection Approach to Surgery. J Clin Oncol. 32(8): 774–782.  

9. Arney, K. (2018). Improving brain-cancer therapies through mathematical modelling. Nature. 561(7724):S52-S53. 

10. Aerts, H. et al. (2018). Modeling Brain Dynamics in Brain Tumor Patients Using the Virtual Brain. Eneuro. 5(3). 

11. Chen, S. et al. (2018). Characterizing Glioblastoma Heterogeneity via Single-Cell Receptor Quantification. Bioeng Biotechnol.  

12. Lê, M. et al. (2016). Personalized Radiotherapy Planning Based on a Computational Tumor Growth Model. IEEE Transactions on Medical Imaging, Institute of Electrical and Electronics Engineers, 11.

13. https://news.mit.edu/2018/artificial-intelligence-model-learns-patient-data-cancer-treatment-less-toxic-0810

14. Yauney, G. & Shah, P. (nd). Reinforcement Learning with Action-Derived Rewards for Chemotherapy and Clinical Trial Dosing Regimen Selection. 

15. https://www.clinicalinformaticsnews.com/2017/09/29/the-intelligent-trial-ai-comes-to-clinical-trials.aspx

Leave a Reply

Be the first to reply using the above form.