Algorithm 1.

Organoid Segmentation

1. Median filter was applied to remove noise (medfilt2, kernel size: $25\times 25\text{\hspace{0.17em}\hspace{0.17em}pixels}$).

2. Gaussian filter was applied to estimate the background, and this background was subtracted from the image (imgaussfilt, kernel size: $450\times 450\text{\hspace{0.17em}\hspace{0.17em}pixels}$).

3. Local standard deviation filter was applied to enhance edges (stdfilt, kernel: $13\times 13\text{\hspace{0.17em}\hspace{0.17em}pixels}$).

4. Otsu’s method was used to segment the edge-enhanced image into three classes (imquantize, multithresh, and number of threshold values: (2). The lowest class, corresponding to the non-organoid regions, was excluded.

5. Segmented regions less than 100 pixels were removed.

6. Holes in segmented regions were filled in (imfill, option: “holes”).

7. Segmented regions were eroded with a disk structure element (imerode, strel, options: “disk,” 9 pixels).

8. Active contours refined the segmented regions (activecontours, options: “Chan-Vese,” 200 iterations, $-0.6$ contraction bias).

9. Separation of touching organoids in the mask was achieved via the standard h-Minima-watershed transform algorithm.

10. Holes in segmented regions were filled-in (imfill, option: “holes”).

11. Edges in segmented regions were smoothed with a Gaussian filter (imgaussfilt, kernel size: $5\times 5\text{\hspace{0.17em}\hspace{0.17em}pixels}$).

12. Segmented regions corresponding to PCOs were filtered based on morphology (regionprops, ismember, options: regions with area $>1000\text{\hspace{0.17em}\hspace{0.17em}pixels}$ and circularity $>0.4$).56