Digital Dissection

Brain Mapping & Brain Image Segmentation

Understanding the anatomic variability of the brain in the population is essential for connecting brain function to brain anatomy and for a deeper understanding of development, aging, and disease. The need of neurosurgeons to visualize the complex central nervous system (CNS) is pushing the use of a more accurate segmentation of different anatomical and pathological structures. While neuroanatomists debate the exact boundary of relatively simpler brain parts and traditional brain atlases identify regions only by pointing to the middle, leaving the interfaces between regions unspecified, the International Consortium for Brain Mapping (ICBM) seeks to create a so-called probabilistic human brain atlas. The technology is so advanced that the model will be able to predict the probability of a particular voxel, in such an interface between regions, being a part of either of the adjoining brain regions. (Voxel in a 3D structure is equivalent to pixel in a 2D image).

The inclusion of physiopathological data from functional MRI (fMRI) provides resolution of a few hundred microns over limited volumes and will identify areas with increased biological aggressiveness within a certain lesion prior to surgery or biopsy procedures. Sectioning, staining and optical digitization of cadaver brains allow even finer spatial and chemical resolution in limited numbers of brains. Such new data acquisition technologies with the concepts of 3D stereotaxic mapping will help create probabilistic maps at a very fine scale with the help of powerful computational tools for analysing high-resolution three-dimensional (3D) brain images. Fully automated neuroanatomic segmentation in large numbers of MRI data sets is essential if questions of normal population variability, normal longitudinal development, and detection of abnormality in single subjects or in groups are to be answered definitively.

Image segmentation is a multi step process of dividing an image into a set of semantically meaningful, homogeneous, and nonoverlapping regions of similar features such as intensity, depth, color, or texture. Subsequent measurement and visualization of anatomical & pathological structures will help in computer aided diagnosis, surgical planning and image-guided interventions.  

Brain pathology visualization

Applications

  • In degenerative diseases like Huntington’s disease and Alzheimer’s disease, the sulci become more open and the ventricles become enlarged. Measurements of these changes can lead to early diagnosis and treatment.
  • In Multiple Sclerosis a significant difference in the overall volume & distribution of affected brain tissue between drug and placebo groups can be used as a measure of drug efficacy.
  • Structural analysis of brain can also help to predict outcomes of Traumatic Brain Injuries.
  • On the same principles of brain volumetric analysis, breast volumetric analysis  can be used for aesthetic planning in breast reconstruction.
  • To visualize and store information in Forensic Case analysis.
  • In addition to direct hypothesis testing, the stereotaxic approach of brain mapping may allow for the detection of unsuspected patterns of interaction among normal brain elements and the isolation of constellations of measurements that characterize specific disease states.
Analysis of changes in the Hippocampus anatomy can be used for an early diagnosis of Alzheimer’s disease.

Challenges

  • Such high technology tools often demand that scientific questions be restated and made more amenable to quantitative analysis.
  • Consent from patients for data collection & utilization and ensuring privacy are important issues which can be tackled by blockchain technology which gives direct ownership of images to patients.
  • Curated medical imaging data collection isn’t an easy job either. Curating medical imaging data includes but is not limited to data anonymization, checking the representative of the data, unification of data formats, minimizing noise of the data, annotation, and creation of structured metadata such as clinical data associated with imaging data.
  • The image segmentation methodology isnt fixed. Research has shown that different segmentation methodologies are better for different structures or pathologies. Validation and quantitative comparison of different segmentation methods is a general problem in medical image analysis. It requires a “ground truth” or gold standard to which the outcome of the segmentation method can be compared. Unfortunately, the “ground truth” does not exist for the analysis of in vivo acquired data in humans. Two methods are being used to establish this ground truth. One is a method by which radiologists review images to make labels and the other is a method based on radiologic reports. In the former case, a lot of time and effort is required and there may be an inter-reader disagreement on the label. In the latter case, the correctness of the labels itself may be unsatisfactory.
  • Incorporation of nonimaging parameters such as behavioral variables, demographic information, and genetic data into the statistical models.
  • Applying deep learning to this data comes with its own problems like absence of an audit trail to explain the decisions of the deep learning model, overfitting and adversarial data impact.

Present & Future

Will brain mapping help speed up clinical trials in case of Multiple Sclerosis? Will an Alzheimer’s patient get an earlier diagnosis so that he has more time to get his personal affairs in order? Will the family of a Road Traffic Accident patient get more concrete answers about their loved one’s chances to pull through? This technology like any other will only make economic sense if it creates significant value in people’s lives.

NeuroShield™, a cloud based Clinical Decision Support Tool for Neurological Disorders, is being developed by InMed Prognostics, a Pune based healthcare tech startup. It provides volume analysis of brain parts segmented from 3D MRI and compares it to Indian reference ranges built specifically from Indian population data.

If Google Maps is showing routes to autonomous vehicles today, the idea of ‘bio-tagging’ target voxels for robotic surgeries with the help of such readily available brain maps doesn’t seem too far fetched. Neither does the idea of two neurosurgeons sitting in a cafe and brainstorming the surgical approach to a holographic reconstruction of a segmented brain pathology seem too futuristic.

SentiAR provides real-time holographic visualization of the patient’s actual anatomy in the clinical setting, literally floating over the patient.