Classify Cells Based on Their Shape

Use convexity defects to classify cells

Start with an image of cells:

Image from Liu, Y., Aubrey, W., Martin, K., Sparkes, A., Lu, C. and King, R.D., 2011. The analysis of yeast cell morphology features in exponential and stationary Phase. Journal of Biological Systems, 19(04), pp. 561–575.

Compute a segmentation mask:

In[1]:=

mask=FillingTransform@Dilation[Binarize[RidgeFilter[cells,1],.01],DiskMatrix[3]]

Out[1]=

Find, measure and highlight contours around cells:

In[2]:=

contours=ComponentMeasurements[mask,"PerimeterPositions",#Count>500&]〚All,2,1〛;

In[3]:=

HighlightImage[cells,{AbsolutePointSize[3],contours},ImageSizeMedium]

Out[3]=

Build a small utility to get the

ConvexHullMesh

of the contours and compute the defects. Using

PeakDetect

, find the junction points between two cells:

In[4]:=

convexityDefects[contour_,t_:2]:=Module[{convexHull,nearest,dist,defects,dents,peaks,peakCount,shape},convexHull=ConvexHullMesh[contour];dist=RegionDistance[RegionBoundary@convexHull][contour];defects=UnitStep[dist-t];peaks=PeakDetect[dist,2,Automatic,t];peakCount=Count[Split[peaks],{1..}];shape=Switch[peakCount,0|1,"Simple",2,"Pair",_,"Complex"];AssociationThread[{"Contour","ConvexHull","Distance","Defects","Peaks","Shape"},{contour,convexHull,dist,defects,peaks,shape}]]

In[5]:=

convexityDefects[contours〚4〛]//Dataset

Out[5]=

It is now possible to construct the lobed cells finder using contour convexity defects and

FindPeaks

In[6]:=

defectsGraphics[contour_Association]:=Graphics[{Lighter[Blue,.8],MeshPrimitives[contour["ConvexHull"],2],Black,Line[contour["Contour"]],Red,AbsolutePointSize[6],Point[Pick[contour["Contour"],contour["Peaks"],1]]}]

In[7]:=

defectsGraphics[convexityDefects[contours〚4〛]]

Out[7]=

In[8]:=

distancePlot[contour_Association,opts___]:=ListLinePlotcontour["Distance"],

,EpilogTextcontour["Shape"],

,PrologInset



Finally, you get the cell shape classifier based on convexity defects, which reports whether the cell is simple (no or one defect), a pair of cells (two defects) or complex (multiple defects):

In[9]:=

GraphicsGridPartition[Table[distancePlot[convexityDefects[contours〚i〛,2]],{i,{5,10,19,26}}],2],





Out[9]=

Publisher Information

Contributed by: Wolfram Staff

Classify Cells Based on Their Shape

Related Symbols

Publisher Information