Mini-Symposia 3

Caroline Chung, MD, The University of Texas MD Anderson Cancer Center
Chief Data Officer and Associate Professor, Department of Radiation Oncology

Medical data is growing in volume and complexity faster than ever before, raising the need for and the promise of benefiting from data science approaches to gain insights from our data to benefit our patients. Across a patient’s journey, multimodal, longitudinal data is acquired and emerging data science methods are starting to tackle time-course multimodal data with an aim to inform treatment decisions and discovery of new and improved therapeutic delivery. There are ongoing pursuits to integrate multimodal data into digital twins to guide patient care and to build out an operational learning health system to improve outcomes for every patient.

However, the sobering reality is that while the volume of data is growing at a pace that is overwhelming the medical system and clinical teams, the proportion of data that is accessible and effectively usable remains limited. Amidst the enthusiasm of artificial intelligence, there is growing awareness and realization that the context of data is critical in the appropriate and meaningful use of data. While currently available big data can be leveraged for some insights, they often have insufficient context to evaluate data quality, biases and appropriate utilization. Leveraging and enriching the metadata will help provide the context to inform data provenance that can also inform data quality and data lineage, which in turn can help ensure appropriate governance of access to data, appropriate use of data for the intended goals and appropriate interpretation of the insights. This foundational change to how data is generated and flowed will require dedicated investment and broad engagement and coordination across the community, but has great potential to elevate the value of data science in oncology and accelerate discovery and clinical translation of these discoveries to impact patients.

Jan-Jakob Sonke, PhD, The Netherlands Cancer Institute
Professor, Adaptive Radiotherapy

Radiation therapy (RT) is a mainstay of cancer treatment. For about 50% of cancer patients, RT is applied during the course of their disease. RT is labor intensive, where the treatment preparation phase including multi-modal medical imaging, definition of targets and surrounding organs at risk and optimization of the treatment plans takes about a week involving a large multi-disciplinary team. Subsequently, treatment is delivered over 3-35 daily fractions depending on the stage and location of the disease. To guarantee that the tumor is irradiated as intended in the presence of daily anatomical changes, generous safety margins are traditionally applied around to target also exposing surround tissue to high levels of radiations.

Data science and artificial intelligence are extensively being studied and increasingly used to improve the efficiency and efficacy of RT. Through auto-segmentation and auto-planning techniques, the treatment preparation step may be reduced to less than 1 hour and ultimately take place directly on the treatment machine in a few minutes. Automatically adapting existing treatments to the anatomy of the day will reduces dose to organs at risk and associated toxicity. Accelerated imaging techniques through learned reconstruction algorithms for in room imaging systems will simultaneously improve image quality and reduce treatment time slots. Large scale dose accumulation provides accurate estimates of the delivered dose and allows detailed modeling of personalized dose-effect relationships. Large scale collection and analysis of (imaging) biomarkers of response will facilitate the design of personalized spatio-temporal modulated treatment regiments.

Large scale privacy compliant data sharing, public databases and/or distributed learning strategies would accelerate data science and artificial intelligence in RT. Capturing high quality longitudinal response biomarkers would accelerate personalized adaptive RT.

Iwan Pouluci, PhD
The University of Texas MD Anderson Cancer Center
Instructor, Department of Interventional Radiology

Interventional Oncology (IO), a branch of Interventional Radiology, is being established as a fourth pillar of cancer care next to medical, surgical and radiation oncology. The field offers local curative treatments such as thermal ablation and palliative treatments such as chemo and radioembolization, or cryoablation of bone metastases among others. In addition, tissue biopsies that provide information about the underlying tumor biology, such as genomic mutations, are the most commonly performed procedure and an integral part of modern cancer care. The hallmark of IO is its reliance on image guidance by CT, MRI, US, and fluoroscopy to perform percutaneous or endovascular minimally invasive procedures. Therefore, IO is a very data rich specialty where each step of the procedure is documented with imaging. This makes IO a premier field to implement data science and AI applications.

The potentially largest impacts on IO themselves will come from automated image processing, such as image segmentation, and outcome predictions to aid in patient selection. Already, AI has enabled intra-procedural assessment of the technical treatment endpoints in thermal ablation and embolization of liver tumors, for which optimum thresholds were identified in large scale data analyses and are being verified on prospective trials. Without these automated image processing and fusion techniques, intra-procedural assessment would not be feasible as it would drastically prolong the procedure. Another promising application is prognostication of treatment responses and risk of complications following IO procedures. Even within the field of IO there are many variables within the treatments that could be optimized and tailored to each specific patient or even tumor. In hepatic embolization for example, the size and type of the embolic material (plastic vs glass), the coating (bland, chemotherapy drugs, or radioactive substances), and the dose are just some of the parameters that could be personalized to deliver an optimum tumor response.

The information contained in this e-mail message may be privileged, confidential, and/or protected from disclosure. This e-mail message may contain protected health information (PHI); dissemination of PHI should comply with applicable federal and state laws. If you are not the intended recipient, or an authorized representative of the intended recipient, any further review, disclosure, use, dissemination, distribution, or copying of this message or any attachment (or the information contained therein) is strictly prohibited. If you think that you have received this e-mail message in error, please notify the sender by return e-mail and delete all references to it and its contents from your systems.

Jon Heiselman, Ph.D.
Research Associate | Department of Surgery |
Memorial Sloan Kettering Cancer Center | Sloan Kettering Institute
Adjunct Assistant Professor | Department of Biomedical Engineering | Vanderbilt University

For many cancer types, surgery has long been regarded as a critical cornerstone of curative treatment. Yet, surgery is a highly dynamic process subject to substantial variations contingent upon the specific surgery type, individual patient-related factors, and preferences of the surgical team. Due to these variations, the reliability of computational approaches to measure and interpret the surgical environment is of central concern to surgical data science. While most operating rooms in current clinical practice lack concordant infrastructure and instrumentation to support intraoperative data acquisition and curation of datasets on the scale where machine learning has already found major success [1], the future of surgical data science will rely on advances in intraoperative instrumentation to drive forth new capabilities in surgical scene sensing, task planning, treatment delivery, and intraoperative assessment. Comprehensive approaches for advancing treatment precision using information extracted under limited visibility within surgical environment will also require simultaneous control over uncertainties introduced by the many confounding factors present during surgery when integrating multimodal data across multiple timepoints.

Furthermore, the ability to optimize patient outcomes within surgical oncology pivotally depends on patient selection and the overarching timing of surgical therapy within the course of cancer management. In select cancer types, recent advances in medical and radiation oncology have enabled total neoadjuvant approaches that suggest durable outcomes may be achieved which preserve surgery as a viable downstream therapy [2] and raise questions towards the timing with which surgical therapy should be delivered to maximize curative potential. The role of data science is growing rapidly towards imaging biomarkers and radiomic analyses for non-invasive determination of disease state and outcome prognostication for surgical selection. However, the ability to establish accurate patient-level ground truth remains a barrier for machine learning methods to move from coarse predictors to personalized descriptors that could become more reliable discriminators to indicate surgical treatment. Highly accurate voxel-level correlation of patient-specific imaging features to histopathology specimens will be an important step towards enabling new supervised approaches in surgical data science to narrow the gap of uncertainty in precisely assessing disease, predicting outcomes, and selecting patients for surgery using rich multimodal information.

[1] Maier-Hein L, Eisenmann M, Sarikaya D, März K, Collins T, Malpani A, Fallert J, Feussner H, Giannarou S, Mascagni P, Nakawala H, Park A, Pugh C, Stoyanov D, Vedula SS, Cleary K, Fichtinger G, Forestier G, Gibaud B, Grantcharov T, Hashizume M, Heckmann-Nötzel D, Kenngott HG, Kikinis R, Mündermann L, Navab N, Onogur S, Roß T, Sznitman R, Taylor RH, Tizabi MD, Wagner M, Hager GD, Neumuth T, Padoy N, Collins J, Gockel I, Goedeke J, Hashimoto DA, Joyeux L, Lam K, Leff DR, Madani A, Marcus HJ, Meireles O, Seitel A, Teber D, Ückert F, Müller-Stich BP, Jannin P, Speidel S. Surgical data science – from concepts toward clinical translation. Med Image Anal. 2022 Feb;76:102306. doi: 10.1016/j.media.2021.102306. Epub 2021 Nov 18. PMID: 34879287; PMCID: PMC9135051.

[2] Garcia-Aguilar J, Patil S, Gollub MJ, Kim JK, Yuval JB, Thompson HM, Verheij FS, Omer DM, Lee M, Dunne RF, Marcet J, Cataldo P, Polite B, Herzig DO, Liska D, Oommen S, Friel CM, Ternent C, Coveler AL, Hunt S, Gregory A, Varma MG, Bello BL, Carmichael JC, Krauss J, Gleisner A, Paty PB, Weiser MR, Nash GM, Pappou E, Guillem JG, Temple L, Wei IH, Widmar M, Lin S, Segal NH, Cercek A, Yaeger R, Smith JJ, Goodman KA, Wu AJ, Saltz LB. Organ Preservation in Patients With Rectal Adenocarcinoma Treated With Total Neoadjuvant Therapy. J Clin Oncol. 2022 Aug 10;40(23):2546-2556. doi: 10.1200/JCO.22.00032. Epub 2022 Apr 28. PMID: 35483010; PMCID: PMC9362876.