Neuro-Oncology + Artificial Intelligence | #AI #ML | 📧 [email protected]
I lived at C House (Harvard/MIT hacker house) this summer. I focused on building an AI/ML visualization tool for simulating spatial tumor dynamics via advanced reaction-diffusion models.
I was an AI x Med developer Resident @Live the Residency -C3 Cambridge, MA. In addition to building out this project, I had the opportunity to live alongside fellow college students who had become tech hackers and founders.
this is me & my dry erase rendition of the RAF-MEK-ERK pathway :)
Originally, I had set out to have the foundations of this for the following…
Understanding DIPG Heterogeneity via Reaction-Diffusion Modeling
But, after talking with researchers in the field. I realized an common challenge in accessing the necessary datasets due to the proprietary nature of them.
I switched my approach, so I could use more readily available datasets, still focusing on end stage brain tumors… this time within adult populations (i.e glioblastoma).
^ graphic indicating current barriers encountered within HGG treatment cases.
As an result, the tool integrates sophisticated algorithms and data-driven approaches to model and predict the spatial and temporal evolution of tumors.
The simulations’ heatmap illustrates the concentration of tumor cells varying across an MRI segment.
ReactDiff’s model is based on partial differential equations (PDEs) that describe the concentration of tumor cells over time and space.
The specific equation accounts for parameters that incl. diffusion coefficients, reaction rates, and (eventually) anisotropic diffusion preferential migration along specific tissue structures.
$∂𝑐(x,t)/∂t = 𝑑∇^2𝑐(𝒙,𝑡)+𝜌𝑐(𝒙,𝑡)(1−𝑐(𝒙,𝑡))$
The reaction diffusion equation is incorporated in the model to help simulate tumor cell growth and behaviours.
This model captures the key dynamics of tumor growth, including…