Cytomorphometric feature tables and analysis code for cell-level vs field-level cervical cell classification (CPSMI2025 and SIPaKMeD)

Cytomorphometric feature tables and analysis code
Cell-Level and Field-Level Cytomorphometric Representation for Cervical Cell Classification: A Cross-Dataset Study of CPSMI2025 and SIPaKMeD
This archive is the reproducibility material for the manuscript above. It contains the
derived cytomorphometric feature tables, the extraction code that produced them,
the complete analysis code, the serialized results behind every table and figure, and the
figures as published.
It does not contain any source images. Both image collections are publicly available
from their original providers (see Source data below) and must be obtained there.
---
Contents
```
features/   cpsmi2025\_cell\_level\_features.csv     2126 rows x 80 cols
            cpsmi2025\_field\_level\_features.csv    1951 rows x 377 cols
            sipakmed\_cell\_level\_features.csv      4037 rows x 80 cols
code/       extract\_cyto\_features.py   produces the tables above from the source images
            cyto\_common.py             loading, feature families, paired matching, grouping
            cyto\_v2.py                 the full analysis protocol (E0-E8)
            verify.py                  independent recomputation of every reported value
            claim\_audit.py             traces each number in the manuscript back to results
            cyto\_figs.py               figure generation (300 DPI)
results/    v2.json, cyto\_summary.json, cyto\_meta.json
figures/    fig1 - fig7 as embedded in the manuscript
```
Source data
SIPaKMeD — https://www.cs.uoi.gr/~marina/sipakmed.html
4049 isolated cells manually cropped from 966 cluster images, five categories.
The providers require citation of:
M. E. Plissiti, P. Dimitrakopoulos, G. Sfikas, C. Nikou, O. Krikoni, A. Charchanti,
"SIPAKMED: A new dataset for feature and image based classification of normal and
pathological cervical cells in Pap smear images," IEEE ICIP 2018, Athens, Greece.
CPSMI2025 — https://data.mendeley.com/datasets/rnfyhrtfkw/1
DOI: 10.17632/rnfyhrtfkw.1
Conventional Pap smear microscopy images, nine subcategories.
Column conventions
`image\_path` is relative to the root of the corresponding source collection, so each
row can be traced to its source image after download. No absolute or machine-specific
paths are included.
`source\_image\_group` is the unit of statistical independence:
SIPaKMeD: the source cluster image (`class/NNN`), recovered from the `NNN\_MM.bmp`
naming convention. 4037 cells resolve to 966 groups (mean 4.18 cells per group).
Cells from one group are not independent; all cross-validation, bootstrapping and
permutation analysis must group by this column.
CPSMI2025: one row per source image, so group = image.
The 74 cytomorphometric descriptors are identical in name and semantics across both
corpora. The field-level table holds the same 74 base descriptors summarised by five
aggregation statistics (mean, std, p25, p50, p75) plus `n\_cells`, the detected-cell count.
Reproducing the analysis
```bash
pip install numpy pandas scikit-image opencv-python-headless scipy scikit-learn matplotlib seaborn
cd code
python cyto\_v2.py e0
python cyto\_v2.py e1 cpsmi\_cell        # also: cpsmi\_field, cpsmi\_field\_mean,
                                       #       cpsmi\_field\_noN, sipakmed\_cell
python cyto\_v2.py e2
python cyto\_v2.py e3
python cyto\_v2.py e4  <arm>
python cyto\_v2.py e56 <arm>
python cyto\_v2.py e7
python cyto\_v2.py e8
python verify.py
python cyto\_figs.py
```
To regenerate the feature tables from the source images after downloading them:
```bash
python extract\_cyto\_features.py --root /path/to/SIPaKMeD  --dataset sipakmed --mode cell  --out sipakmed\_cell\_level\_features.csv
python extract\_cyto\_features.py --root /path/to/CPSMI2025 --dataset cpsmi    --mode cell  --out cpsmi2025\_cell\_level\_features.csv
python extract\_cyto\_features.py --root /path/to/CPSMI2025 --dataset cpsmi    --mode field --out cpsmi2025\_field\_level\_features.csv
```
All partitions are deterministic given seed 20260901.
Licence
Code: MIT (see LICENSE).
Derived feature tables: released for research use, subject to the terms of the two source
collections, which must be cited as above.