Novel machine-learning approaches to create a structurally accurate virtual model of the heart cell
Project Leader: Vijayaraghavan Rajagopal
Collaborators: Eric Hansenn (Bio21 Institute), Christian Soeller (University of Exeter)
Primary Contact: Vijay Rajagopal (email@example.com)
Keywords: bioinformatics; cardiovascular disease; cellular geometrics; machine learning; medical image analysis
Disciplines: Biomedical Engineering,Computing and Information Systems,Electrical & Electronic Engineering
Domains: Convergence of engineering and IT with the life sciences
Research Centre: Systems Biology Laboratory
Cardiac cells are called striated muscle cells due to their highly organised internal ultrastructure. New microscopy technologies such as serial-block-face imaging, super-resolution microscopy and electron tomography are generating new information regarding cardiac cell structure from micro to nanometer resolution. These structural arrangements are altered in lab conditions (when culturing them for pharmacological studies,) as well as disease conditions (such as heart failure). However, a systematic, in-depth quantification of differences within a cell has not been attempted before.
This is largely due to: the fact that biologically motivated structural studies typically focus on one or a few components of the cell and therefore do not perform a cell-wide structural analysis; and simultaneous visualisation of all components of one heart cell is technically difficult because of limitations in microscopy methods.
The aims of this project are: To develop novel machine-learning algorithms to segment cardiac cell structural components from 3D microscopy data; and to generate 3D visualisations of cardiac cell organisation in healthy and disease conditions using statistical measurements of cell organisation derived from the segmentation process.
(1) Develop a seamless, automated method to segment and measure the spatial distribution of ion-channels in immuno-labelled confocal and super-resolution images of heart cells. Use statistical and machine-learning approaches to quantify and visualise these distributions in a virtual model of the cell. Initial targets – RyRs, SERCA, NCX, IP3Rs, LCC.
(2) Develop machine-learning approaches to semi-automatically segment serial-block-face imaging data of cardiac cell mitochondria and myofibribls.
(3) Develop algorithms to segment and analyse two key tubular networks -t-tubules and sarcoplasmic reticulum-to generate graphical network models of these critical components of the heart cell.
(4) Generate a 3D visualisation model of the heart cell that fuses information from the different microscopy datasets using steps 1 to 3.
(5)cAs a stretch-goal, apply these techniques to structural data on cells from diabetic or hypertrophic (cells enlarged) animal models (rat/mice).
The computer model of the heart cell will be the first structurally accurate 3D visualisation of its critical components. The virtual reality model will be used by the computational modeling community for simulating and visualising cardiac cell function in health and disease. Methods for machine-learning and measuring cellular structural organisation will be of use to the wider cell-biology community. The ability to seamlessly analyse microscopy data to build virtual models will help many scientists rapidly build 3D models of other cell types.