Capsular contracture (CC) is an ailing complication that arises commonly amonst breast cancer patients after reconstructive breast implant surgery. CC patients suffer from aesthetic deformation, pain, and in rare cases, they may develop anaplastic large cell lymphoma (ALCL), a type of cancer of the immune system. The mechanism of CC is unknown, and there are few objective assessments of CC based on histology.
Figure 1: Baker grade
Baker grade is a subjective, clinical evaluation for the extent of CC (See Fig 1). Many researchers have measured histological properties in CC tissue samples, and correlated theses findings to their assigned Baker grade. It has been found that a high density of immune cells is associated with higher Baker grade.
These immune cells include fibroblasts and myofibroblasts, which can distort surrounding tissues by contracting and pulling on them. The transition from the fibroblast to myofibroblast phenotype is an important driving step in many fibrotic processes including capsular contracture. In wound healing processes, the contactility of myofibroblasts is essential in facilitating tissue remodelling, however, an exess amount of contratile forces creates a positive feedback loop, leading to the formation of pathological capsules with high density and extent of deformation.
Myofibroblasts, considered as an “activated” form of fibroblasts, is identified by the expression of alpha-smooth muscle actin (\(\alpha\)-SMA). However, this binary classification system does not capture the full range of complexities involved in the transition between these two phenotypes. Therefore, it is beneficial to develop a finer classification system of myofibroblasts to explain various levels of forces they can generate. One recent work uses pre-defined morphological features of cells, including perimeter and circularity, to create a continuous spectrum of myofibroblast activation (Hillsley et al. 2022).
Research suggests that mechanical strain induces change in cell morphology, inducing round cells that are lacking in stress fibers into more broad, elongated shapes. We hypothesize that cell shapes influence their ability to generate forces via mechanisms of cell-matrix adheshion and cell traction. Further, we hypothesis that cell shape is directly correlated with the severity of CC by increasing contractile forces.
In order to test these hypothesis, we will take a 2-step approach. The first step involves statistical analysis on correlation between cell shapes and their associated Baker grade. To do this, we collect cell images from CC samples with various Baker grades, using Geomstat we can compute a characteristic mean cell shape for each sample. Then, we cluster these characteristic cell shapes into 4 groups, and observe the extent of overlap between this classification and the Baker grade. We choose the elastic metric, associated with its geodesic distances, since it allows us to not only looking at classification, but also how cell shape deforms. If we can find a correlation, the second step is then to go back to in-vitro studies of fibroblasts, and answer the question: can the shapes of cells predict their disposition to developing into a highly contractile phenotype (linked to more severe CC)? I don’t have a concrete plan for this second step yet, however, it motivates this project as it may suggest a way to predict clinical outcomes based on pre-operative patient assessment.
Cell segmentation
I was provided with histological images of CC tissues, by a group in Coppenhagen {add credit and citations}. The images are \(\alpha\)-SMA stained in order to visualize myofibroblasts, each image is associated with a Baker Grade and the age of the implant. The first step is to preprocess the images and segment the cells. Fiji is a great tool for this purpose.
TO ADD: details on training the classifier.
Pre-Processing
Sort labelling data
The segmentation data can be exported as a file containing 2D coordinates of all pixels that are marked as borders. First, we need to identify individual cells from this data. We may view pixels as nodes in a graph, the problem then becomes splitting an unconnected graph into connected components. A tricky part is to process cells with overlapping/connected borders. > TO ADD: details on this algorithm.
From here, a few simple bash commands allow us to import the resulting data files as a numpy array of 2D coordinates, as an acceptable input for GeomStats.
# replace delimiters with sed
sed -i 's/],/\n/g' *
sed -i 's/,/ /g' *
# remove [ with sed
sed -i 's|[[]||g' *
import sysfrom pathlib import Pathimport numpy as npfrom decimal import Decimalimport matplotlib.pyplot as plt# sys.prefix = '/home/uki/Desktop/blog/posts/capsular-contracture/.venv'# sys.executable = '/home/uki/Desktop/blog/posts/capsular-contracture/.venv/bin/python'sys.path=['', '/opt/petsc/linux-c-opt/lib', '/home/uki/Desktop/blog/posts/capsular-contracture', '/usr/lib/python312.zip', '/usr/lib/python3.12', '/usr/lib/python3.12/lib-dynload', '/home/uki/Desktop/blog/posts/capsular-contracture/.venv/lib/python3.12/site-packages']directory = Path('/home/uki/Desktop/blog/posts/capsular-contracture/cells')file_iterator = directory.iterdir()cells = []for filename in file_iterator:withopen(filename) asfile: cell = np.loadtxt(file, dtype=int) cells.append(cell)print(f"Total number of cells : {len(cells)}")
Total number of cells : 3
Since the data is unordered, we need to sort the coordinates in order to visualize cell shapes.
def sort_coordinates(list_of_xy_coords): cx, cy = list_of_xy_coords.mean(0) x, y = list_of_xy_coords.T angles = np.arctan2(x-cx, y-cy) indices = np.argsort(angles)return list_of_xy_coords[indices]
sorted_cells = []for cell in cells: sorted_cells.append(sort_coordinates(cell))
Problems: 1. Some weird shape are reading weird (potentially wrong?) 2. Need some kind of bash or python script to process labelling data into numpy arrays 3. Labelling coordinates comes in a clump, can’t think of a easy way to process overlaps of cells (might discard) 4. Cell shapes are in 2D, lose information
References
Hillsley, Alexander, Matthew S Santoso, Sean M Engels, Kathleen N Halwachs, Lydia M Contreras, and Adrianne M Rosales. 2022. “A Strategy to Quantify Myofibroblast Activation on a Continuous Spectrum.”Scientific Reports 12 (1): 12239.
