Damage in brain tissue follows three significant time scale. In the short time scale (miliseconds, seconds) it is traumatic brain injury, in the intermediate time scale (minutes, hours) it is the damage induced by surgical tools, in the long time scale (months, years) it is the brain degeneration.  In our lab, we are developing mathematical models (combining physics and data driven approaches) of soft brain tissue to incorporate these three distinct damage evolution. The mechanical property estimation of brain tissue non-invasively  is a challenge. MRI elastography has came up as a new tool to non-invasively estimate the material properties of brain tissue. MRI image processing along with, inverse estimation of material properties is the basic building block of elastography technique.   We thrive to develop robust algorithm to make this MRI elastogrphy technique accessible to the medical community in India. 

Magnetic Resonance Elastography (MRE): A noninvasive digital sense of touch in Medical Imaging

Pathological conditions often alter the stiffness or elasticity of tissues. For example, tumors are generally stiffer than the surrounding healthy tissue. This change in stiffness can be an important diagnostic marker. While hand or manual palpation  has traditionally been the primary noninvasive clinical technique for detecting tissue abnormalities. It involves feeling the tissue’s texture, firmness and consistency and making a qualitative assessment using physician’s bare hands. Naturally, its effectiveness heavily relies on the physician’s experience, skill and training.

Modern medicine has developed several advanced non-invasive imaging techniques to measure tissue stiffness in a more objective and quantifiable manner. Ultrasound elastography, MRI elastography are two of the most common names. Unlike manual palpation, these imaging techniques provide quantitative data, allowing for more accurate and consistent assessments and even being able to assess deep tissue like brain, which are not assessable through manual palpation. Elastogram and images can also provide time series information; monitoring, visualizing disease progression over time. These techniques still require skilled operators but the advanced AL/ML algorithms can make it much less reliant on practitioner's subjective interpretation.  These imaging techniques have applications beyond detecting tumors. They can be used to diagnose and monitor various conditions, including liver fibrosis, breast lesions, prostate cancer, and musculoskeletal disorders. The in-vivo tissue physics can be incorporated into virtual surgery model to evaluate tissue-device interaction for pre surgical planning, surgical guidance.

In MRE, sequential images of shear wave motion within the tissue, induced by internal or external stimuli using Magnetic Resonance Imaging (MRI), are recorded. This kinematical information is then utilized in an inversion algorithm based on the principles of elastodynamics to reconstruct an elasticity map of the area of interest. Figure 1 illustrates two cases of soft (top row) and stiffer (bottom row) meningioma, the most common type of brain tumor. The T1-weighted anatomical images on the left column confirm the presence and location of the tumor only. However, it is the stiffness map obtained through MRE (displayed in the right column) for each case that provides a quantitative assessment of their stiffness, further confirming their presence within the brain. This quantitative evaluation of tumors to be resected aids in effective planning of the surgical procedures.

The inversion algorithm for estimating the tissue stiffness involves solving an inverse problem. The objective is to deduce the unknown material parameter field based on the available experimental measurements. The inverse problem is posed as an optimization problem subjected to the PDEs governing the wave motion in elastic or viscoelastic tissues. The inverse problem is often solved iteratively using large-scale optimization algorithms. Moreover, machine learning presents a promising avenue, either as a standalone solver to find the optimum solution or as an assisting tool for learning the unknown parameter field using the provided dataset.

Objective: The current project aims to develop AI/ML, differentiable physics based inversion algorithms for MRI based elastogram construction for neuroimaging applications.  A secondary objective is to create a tissue physics informed virtual surgical framework for neuro-surgical simulation. 

Research Area: Foundational Model for Multi-modal,  scientific machine learning based predictive dynamics of disease progress.

Background Machine learning algorithms have made remarkable progress in data driven healthcare domain, offering various ways of extracting features from labeled data to diagnose, classify illnesses with great accuracy. These methods include unstructured data in the form of texts like prescriptions and patient descriptions, in the form of images such as MRI scans, CT scans, PET scans etc. as well as structured data like blood reports, pathological test report etc. The challenge is to bring all these different types of data together to create a mechanistic/digital twin type model that can help not only diagnose diseases both early and accurately, for e.g, early diagnosis of moya moya (cerebral stroke), diabetes, dementia, Alzheimer, cancer etc, but predict the progression of chronic disease over a period of time also (predictive medicine). It will help in managing the progression of chronic disease, design personalized treatment, continuous monitoring of psychiatric diseases and maintaining sustained remission if not complete cure.   

Objective:  The objective of the present research is to develop ML models fusing multimodal (images, report, time series etc) historical data with the bio-physical/bio-mechanical understanding of disease for predictive medicine. We plan to combine data driven ML model with computational physics driven disease model for  explainability and generalization.

Novelty: There are lot of innovative developments in application of machine learning in disease diagnosis. However, current methodologies often lag in harnessing the full potential of these advancements, particularly in case of multimodal medical datasets. Our present research aims  to address this disparity by using state-of-the-art diagnostic techniques and enhancing our model through the integration of advanced feature extraction methods and generative machine learning algorithms.

End product: Building an enhanced ML model which includes important features of different modalities for future prediction of disease progression and management. 

Research Area: Digital Twin of Human; Foundational Framework for Human-in-loop design of Intelligent Soft Robotic Wearables for Augmentation 

https://sites.google.com/view/jatc-nse/home

Symbiotic biomechanics is a field to study, analyze how man-machine interface in wearable robotics affect the human body natural biomechanics. The word symbiosis comes from co-existence, were device and human body interect. 

Soft exosuit has many applications, wheather it is at subzero temperature where muscle activity reduces, manning steep, rough terrain, or trekking up stiff slopes to avoid enemy surveillance, a soft robotic ExoSuit can find many applications, for e.g.,  protecting from injuries, reducing fatigue etc. An EMG/EEG and a combination of other non-invasive sensor suit controlled wearable soft ExoSuit has many advantages over its bulky, hard exoskeleton counterpart. On top of that, the materials used in the manufacturing are mostly polymeric material, textile composite-hence it, coupled with modern rapid prototyping techniques, makes the device form factor easy to build, cheap and affordable. ExoSuit, being wearable by nature, fits well with the definition of collaborative robotics (Cobotics). Other than augmentation, a soft ExoSuit also has applications in assistive motor rehabilitation purposes for a stroke affected patients and fits well into the medical robotics vertical. ExoSuit can also act as fatigue and injury, reducing augmentation device for factory assembly line workers.