Quantitation of nuclear atypia


Hi, I was wondering how we can use Cellprofiler to assess nuclear atypia. Nuclear atypia is a feature of malignancy and it usually means several morphological characteristics of nuclei such as: irregular shape (as oppsed to perfect round or ellipse), enlarged size, hyperchromasia (darker than usual), clumpy chromatin (having dark and light areas in a single nucleus), disorganization (nuclei pointing to different directions), etc. Size and darkness (intensity) can be easily measured on Cellprofiler. But is there anyway to assess the irregularity or symmetry of nuclei, and heterogeneity of intensity? I guess disorganization can be assessed by comparing the long axis of one nuclei against those of its neighbors, and see if they are parallel. Any suggestions? Thanks a lot.



I think CellProfiler can make the measurements you are looking for. Asymmetry/irregularity of nuclei can perhaps be quantified using Eccentricity, one of the features measured by MeasureObjectAreaShape. The intensity features you are looking for (dark and light spots, etc) can probably be quantified using the features in MeasureTexture (see the help for the MeasureTexture module for a description of the features). If this isn’t totally what you were looking for, a new module in the upcoming release measures radial distribution of a stain, which could also be helpful for quantifying blotchy nuclei.



And, it sounds like this project would benefit substantially from using the machine-learning function called “Classify” within CellProfiler Analyst. Machine-learning is used to quantify phenotypes that are a combination of several features of the cells. So, instead of you choosing one or two measurements by hand that reflect your phenotype, you would sort some example cell images as ‘yes’ or ‘no’ for your phenotype and allow CellProfiler Analyst to decide what CellProfiler-produced measurements can be combined to best represent your phenotype (probably the measurements that Kate mentions would be highly used by the machine learning).

Using CellProfiler Analyst requires that you use the ExportToDatabase module in CellProfiler, and this means you will need to have your local IT people give you access to a MySQL database (open-source). Take a look at the “CellProfiler Analyst Demo movie: Classifying cells with machine learning” and consider using it for your project:



Hi, Kate and Anne, I really appreciate your comments! I am just starting to give a final push to my project and hopefully will write the paper soon. I didn’t realize CP Analyst has a machine learning function. I’ll play with it and may come back with some new questions. Thank you! - Jie