Lab: Diagnosing Cancer

STAT 20: Introduction to Probability and Statistics

Fine needle aspiration biopsy

Often when someone is suspected to have cancer (e.g. a bump is found) a fine needle aspiration biopsy is taken to determine whether or not the growth is cancerous (malignant – may grow dangerously out of control) or not (benign).

Artificial intelligence in medicine

• Automating certain diagnostic tasks can increase access to healthcare

• Global shortage of pathologists, especially outside of wealthy healthcare systems

  • Expert pathologists take years to be fully trained (4 year medical school + 4 year residency)

https://www.linkedin.com/pulse/how-ai-can-help-address-global-shortage-pathologists-colangelo/

Lab 6: breast cancer diagnosis

• Samples are 568 biopsies
  • Each biopsy has 30 features
• Goal: classify biopsy as benign or malignant

The tissue is purple because it is stained with Hematoxylin and Eosin. These are purple and pink stains that make the tissue structure easier to see under a microscope.

Most of the visible objects in these images are cell nuclei, not the full cell; the cytoplasm is mostly invisible on these images.

(From https://stanfordhealthcare.org/medical-conditions/cancer/cancer.html): Tumors can be benign (noncancerous) or malignant (cancerous). Benign tumors tend to grow slowly and do not spread. Malignant tumors can grow rapidly, invade and destroy nearby normal tissues, and spread throughout the body.

Nuclear morphology

• Morphology = what the cell looks like under a microscope
  • size, shape, texture
• Cells in malignant biopsies tend to
  • be larger
  • irregularly shaped
  • highly variable
• Only measure morphology of cell nucleus

10 nuclear morphology features

We compute 10 morphological features for each cell nucleus in the biopsy image.

It may be helpful to first draw one unit (one glass slide, that you would see under the microscope). Then in this slide, you make a data frame with cells, and then group by-summarise. Then the resulting calculations become one row in the final dataset.

30 biopsy features

Each biopsy has 30 features; these come from computing 3 summary statistics (mean, max, standard deviation) of each of the 10 cell features to summarize the population of cells in the biopsy.

diagnosis	radius_mean	area_mean	radius_sd
B	13.700	571.1	0.2431
B	12.720	501.3	0.2954
B	11.750	422.9	0.4384
M	13.440	563.0	0.2385
M	12.450	477.1	0.3345
M	19.590	1214.0	0.7364
B	12.060	448.6	0.1822
M	18.050	1006.0	0.9806
B	8.734	234.3	0.5169
B	13.210	537.9	0.2084
M	15.460	731.3	0.3331
M	14.220	609.9	0.2860
B	11.500	407.4	0.3927
M	14.780	668.3	0.3577
B	9.676	272.5	0.2744
B	12.580	489.0	0.2719
B	9.738	288.5	0.1988
B	10.750	355.3	0.2525
B	11.060	366.5	0.1779
B	12.880	514.3	0.2116
M	15.660	773.5	1.2920
M	23.090	1682.0	1.2910
M	19.450	1169.0	0.5959

25:00