Tracking cancer growth by intelligent displacement biomaps

Brain tumors are one of the deadliest forms of cancer and the primary goal of treatment is simply to decelerate tumor growth. Still, after decades of research, the survival outcome for this patient group has hardly improved. A reason for this paradox is a one-size-fits-all approach to diagnosis and treatment that does not do justice to the time of diagnosis and inherent heterogeneity of the disease. Instead, the patient’s own data and disease on an individual level should help guide the treatment decision making process, so-called personalized medicine. Until now, this approach has not been technically feasible for radiographic images, and Magnetic Resonance imaging (MRI) in particular. Assessing the share number of images acquired during a standard MRI exam constitute an overwhelming task for any human, even for an expert radiologist. The TrackGrowth project takes advantage of artificial intelligence to reach the goal of personalized medicine. By use of state-of-the-art deep learning technology, we will obtain new knowledge that will help stratify patients with brain cancer to receive the best personalized treatment option - at the right time. We present a new paradigm for tumor diagnostics, coined displacement biomaps, where deep learning brings to life hidden information on cancer growth stored in the MRI data. The process of disease growth may involve changes in the tissue microenvironment months or years before it becomes visually apparent to an expert on MRIs. Our preliminary data show how a growing tumor change the entire architecture of the brain more than half a year before these changes are observed by traditional diagnostic means. This new information will allow physicians to make early decisions on treatment – a critical step for the patients in terms of improved quality of life and prolonged survival. To assess the impact of our displacement biomaps in a real clinical setting, we apply our technology to an ongoing clinical trial at Oslo University Hospital aiming to reduce the pressure a brain tumor exerts on its surroundings and thereby improve therapy. The work performed until September 2024 includes analysis of novel biomarkers by MRI, the use of AI for predicting tumor growth, and AI explainability. Key achievements include the publication of findings in prestigious journals, and at conferences, and the integration of the analytical pipelines into clinical workflow. Our software prototype is now integrated in a hospital-approved framework within the picture archiving and communication system (PACS). The technology is protected by an ongoing patent (covering intended use) currently in national phase in the US and Europe (Inven2 commercialization project 21091). We are now demonstrating the clinical value of the software in a realistic setting. We have established a quality management system (QMS) to ensure technical development in accordance with best practices and applicable regulations and set requirements for the commercial product. With this system, the development of the TrackGrowth code is refactored, appropriately documented, and packaged in a containerized format. This will allow for clinical validation for regulatory approval and technical validation that the tool can be deployed through 3rd-party marketplace platforms. Another cornerstone of the TrackGrowth project is the development of a pioneering network designed for predicting glioma growth. This tool, built on diffusion probabilistic models and deep segmentation networks, forecasts tumor evolution and simulates future MRIs for different treatment scenarios, leveraging actual post-surgery longitudinal MRI data. A manuscript detailing these early results, including the generation of synthetic MRIs and assessments of uncertainty are currently undergoing peer-review. Regarding the visualization of tumor progression, the TrackGrowth project has enhanced the interpretability of our deep learning models for survival analysis. Through the examination of saliency maps, the reliability of the AI models has been assessed, demonstrating the association between tumors in vital brain areas and reduced patient survival rates . Early accomplishments also include the exploration of innovative biomarkers combining magnetic resonance elastography (MRE) measurement of tissue stiffness with RNA sequencing of tissue biopsies to explore the molecular characteristics of the stiffness signal. This method has disclosed the molecular attributes of tissue rigidity in glioblastoma, associating increased tissue hardness with the reorganization of the extracellular matrix (stroma) and a decrease in patient lifespan (PMID: 37066109). Finally, the TrackGrowth team has engaged in international collaboration and dissemination activities, notably through its participation in European networks like the GliMR2.0 - COST Action CA18206 project, and the EUCAIM (EUropean Federation for CAncer Images) project under the EU4Health umbrella.

Brain tumors are one of the deadliest forms of cancer and the primary goal of treatment is simply to decelerate tumor growth. Still, after decades of research, the survival outcome for this patient group has hardly improved. A reason for this paradox is a one-size-fits-all approach to diagnosis and treatment that does not do justice to the time of diagnosis and inherent heterogeneity of the disease. The TrackGrowth project takes advantage of state-of-the-art deep learning technology to obtain new knowledge that will help stratify patients with brain cancer to receive the best personalized treatment option - at the right time. We present a new paradigm for tumor diagnostics, coined displacement biomaps, where deep learning brings to life hidden information on cancer growth stored in the imaging data. Our preliminary data show how a growing tumor change the entire architecture of the brain more than half a year before these changes are observed by traditional diagnostic means. This new information will allow physicians to make early decisions on treatment – a critical step for the patients in terms of improved quality of life and prolonged survival. Our network of key stakeholders at Oslo University Hospital (OUH) constitutes a powerhouse for imaging-based diagnostics and artificial intelligence (AI) in clinical oncology. We possess a unique computational infrastructure made ready for AI that allows us to assess the impact of the displacement biomaps directly in the clinical workflow.