Tracing Brain Tumors Through Deep Time

Patients (pediatric and adults) with a diagnosis (radiological/ histopathological) of primary brain tumors registered in the neuro-oncology disease management group between January 2005 and December 2023 will be screened. Approximately 500-600 patients are expected to be eligible per year with imaging data (approximately 500 patients are treated annually with radiation in our center with available CT data for radiation planning, and another 100-200 patients having pre-operative or post-operative scans). For the above-mentioned time period, data is expected to be available from approximately 10,000 patients, which will be the upper limit of sample size for the current study.

The area of the primary tumor (or cavity and residual tumor indicating original location for post-operative data) will be segmented on CT and /or MRI as available. The peritumoral edema will be excluded from the segmented region. The segmentation will be done manually in an initial cohort of approximately 200-500 patients. Subsequently, a machine learning algorithm like a 3D U-net or deep learning-based technique will be trained on the initial data (and validated on the next 100-200 patients to assess algorithm accuracy and robustness) for rapid implementation and segmentation of the large data set. Once brain tumor regions are identified across the entire population, density maps will be generated to reciprocate the location of tumors on a quantitative scale as per age of the patient during diagnosis (age in years as continuous data and categorical data, i.e., age groups, e.g., infants, children, teens, adolescents. adults, and elderly). The generated density maps will be compared with regions of vertebrate brain regions (with openly available literature) across species with regards to the geological scale/ deep time units, e.g., in units of 10-50 million years. Similarly, the skull bony anatomy will be extracted from CT and/ or MRI data (applying techniques like window intensity thresholds without the need for segmentation). Patients with major defects in the calvarial skull from increased intracranial pressure or surgical interventions will be excluded from the analysis of calvarial anthropometry (however, it will be available for skull base anatomy assessment). The organizational patterns will be analyzed using machine learning models and other statistical models like Bayesian statistics and compared with other publicly available normal human populations without brain tumors (adjusting for age, race as applicable), fossil data of vertebrates/ hominids, non-human primates for link recognition. The density maps and anthropometric data will be compared within the entire cohort of patients with brain tumors (from the study) stratified by factors like age (as mentioned earlier), tumor location (e.g., supratentorial vs. infratentorial), tumor grade (benign vs. low grade vs high grade). The statistical analysis for density maps and anthropometry will be done by sharing anonymized data with collaborators with expertise in similar research from the Indian Statistical Institute (Geological Studies Unit and Interdisciplinary Statistical Research Unit).