(D-147) Disease Diagnosis at the Cellular Level using MATLAB-based Image Processing

Introduction:: Disease diagnosis through MatLab-based image processing is a very attractive option for detecting various diseases at the cellular level. PolyFiberQuant is one such MatLab based program was created to analyze the difference in fiber structure between healthy lung cells and lung cells with cystic fibrosis.  A confocal image of a cell is typically processed through the software and a map of the fiber structure is returned with the length of the fibers, the mean length, the standard deviation, the length of the filaments, branch points and endpoints in a 1000um^2 area, the Euler’s number and the Fiberscore of the image. This information is compiled for each image and patterns in the data are analyzed for differences between the fiber structure of healthy and diseased cells. The main purpose of this research project is to test the cells from multiple organs to see if similar patterns exist across different datasets throughout the body. This could be used as a method of diagnosing and identifying diseases at a cellular level. The process is animated and can be processed in less than two seconds as opposed to being mapped by hand. This quickens the process and removes bias when looking at different diseases.

Materials and Methods:: The first step in the analysis of the fiber structure of cells is the collection of confocal images of various cell types in both healthy and diseased variations. Those images are processed through PolyFiberQuant on Matlab. PolyFiberQuant works by applying image filters to smooth the image. A kernel mask is then applied to the image which accesses the brightness of a color on a local level. The pixels that have a higher intensity than the surrounding pixels are marked as possible fibers and the direction of the fiber is calculated by the distribution of the high intensity pixels. The pixels going across a fiber have a gaussian distribution and the pixels along a fiber have a low deviation in intensity. The length of the fiber is calculated using the Kimura method. The fibers are mapped on top of the original image.  The mean length, distribution of the length, the Euler’s number and standard deviation of the length of all the fibers are compared for patterns in the healthy cells or the diseased cells. Differences in the data extracted from the images are marked as a possible identifier of disease in cells.

Results, Conclusions, and Discussions:: Healthy and cancerous breast cells were the first set of cells tested. Healthy breast cells had many fibers that were less than 50 um with a fiber around 200 um and a fiber that was longer. The cancerous breast cells had a fiber that was significantly longer. Each cell had fibers between 400 and 2000 um in length with either no other fibers or only fibers less than 50um. Both lung and bone cells also had very long fibers with  lengths between 600 and 2000 um. The Liver cells were tested as both healthy and diseased with biliary atresia. Both had long and short fibers of a similar distribution. The Healthy cells had fibers that were slightly longer though. The healthy liver cells had a mean length of 148.88 um while the mean length of the cells with biliary atresia  had a mean length of 81.75 um. The mean Euler’s number of the healthy liver cells was -64.5 with the mean Euler’s number of the diseased cells was -75.67.

The patterns of the fiber lengths make sense as with cancer the cell grows abnormally large and the shape becomes malformed. An increase in the length of the cell structure and an increase in overlap of the cell fibers does make sense for cancerous cells. For the liver cells with biliary atresia a lower fiber length makes sense as biliary atresia is a condition involving the narrowing of bile ducts in the liver. This means the shorter fibers in a cell makes sense as the tissue is smaller.  In both the cancerous and the biliary atresia cells the Euler’s number was lower than the cells that were healthy. This pattern in the processing of cells could be used to diagnose diseases on a cellular level.


Acknowledgements (Optional): : We thank Fairfield University for supporting the Summer Undergraduate Research Residency program and for providing the needed computational resources.

References (Optional): :

[1]        S. A. Shah, P. Santago, and B. K. Rubin, “Quantification of biopolymer filament structure,” Ultramicroscopy, vol. 104, no. 3–4, pp. 244–254, Oct. 2005, doi: 10.1016/j.ultramic.2005.04.007.

[2]        R. Murphy, A. Turcott, L. Banuelos, E. Dowey, B. Goodwin, and K. O. Cardinal, “SIMPoly: A Matlab-Based Image Analysis Tool to Measure Electrospun Polymer Scaffold Fiber Diameter,” Tissue Eng. Part C Methods, vol. 26, no. 12, pp. 628–636, Dec. 2020, doi: 10.1089/ten.tec.2020.0304.

[3]        M. O. Henke, G. John, M. Germann, H. Lindemann, and B. K. Rubin, “MUC5AC and MUC5B Mucins Increase in Cystic Fibrosis Airway Secretions during Pulmonary Exacerbation,” Am. J. Respir. Crit. Care Med., vol. 175, no. 8, pp. 816–821, Apr. 2007, doi: 10.1164/rccm.200607-1011OC.