colon & code
01Cold open

What if a CT scan
could map your surgical workspace?

Auto-ISD · open-source CT pelvimetry · IJCARS 2026 · Huang et al.

scroll to begin
02 The problem

Two trained surgeons. Same CT.
A 9 mm spread.

In our blinded validation study, two board-certified colorectal surgeons measured the inter-spinous distance on 70 paired axial CT slices. Their 95% limits of agreement spanned −6.28 to +2.94 mm — a window of more than 9 mm of legitimate disagreement between trained raters.

Inter-rater ICC (R1 vs R2)
0.962
Two-way random, absolute agreement, single measures.
Mean bias (R1 − R2)
−1.67mm
Systematic shift between raters, not random noise.
95% LoA spread
9.22mm
From −6.28 to +2.94 mm. Wider than the ISD range that distinguishes a "narrow" pelvis from a normal one.

At this much human variability, you cannot tell a difficult pelvis from an easy one by manual measurement alone.
And no model built on top of unstable measurements can be trusted.

03 The pipeline

One CT, in. Reproducible numbers, out.

Auto-ISD takes a pelvic CT in NIfTI format and runs a deterministic pipeline: bone segmentation, femur-anchored search window, ISD profile sweep, valley detection, compartment metrics. No manual landmarks. No vendor preprocessing.

~/auto-isd · Patient_011.nii.gz
04 The algorithm

Sweep every slice. Find the narrowest.

For every axial slice in a femur-anchored window, Auto-ISD measures inter-spinous distance, smooths the profile, and selects the slice at the narrowest local minimum. When no clear valley exists, it falls back to the longest plateau.

Sweep every slice. Find the narrowest.

Scan
Loading anatomy
05 · 3D anatomical reconstruction
Case A. A fat-rich pelvis.
ISD 94.4 mm · z=25 · valley
drag to rotate · two-finger pinch on mobile
06 Beyond ISD

ISD is a starting point. Not a verdict.

Inter-spinous distance describes the pelvis at one slice. To predict surgical difficulty, the algorithm also computes the pelvic posterior compartment: the region a TME surgeon actually works in.

Posterior pelvic triangle area

The geometry of the operative field.

Triangle bounded by the bilateral ischial spines and the most anterior point of the sacrum. Captures both the transverse narrowing (ISD) and the anteroposterior depth in a single geometry.

cohort mean 15.25 cm² · auto-vs-manual ICC 0.893
ISD-L ISD-R sacrum ISD
Posterior pelvic fat area (pPFA)

Why two identical bony pelvises aren't.

Patients with the same ISD can present with very different operative conditions. The pelvic fat occupying the triangle reduces the surgeon's actual working space. pPFA quantifies that occupancy directly from torso fat segmentation.

cohort mean 7.62 cm² · contrast robustness ICC 0.87
pPFA fat occupies the triangle
Working space

What's left for the surgeon.

Triangle area minus bowel and fat occupancy. The residual space theoretically available for instrument maneuvering during posterior mesorectal dissection. A composite of geometry and tissue.

cohort mean 5.54 cm² · contrast robustness ICC 0.90
bowel working space
07 Validation

Better than two surgeons agree with each other.

Across 70 cases of blinded manual annotation by two board-certified colorectal surgeons, the automated pipeline's agreement against the manual reference exceeded the agreement between the two raters themselves. Bland–Altman, ICC, and contrast-robustness analyses are in the IJCARS paper.

Auto vs manual · ISD
0.977ICC
Bias 0.45 mm. 95% LoA −3.83 to +4.74 mm. Higher than the inter-rater ICC of 0.962.
Auto vs manual · triangle area
0.893ICC
Three landmarks instead of one accumulate localization error. Still good-to-excellent agreement, no significant systematic bias.
Contrast vs non-contrast · ISD
0.99ICC
69 paired CT acquisitions. Mean bias 0.09 mm. Same patient, contrast or not, same number out.
Pipeline success rate
100%
73 contrast-enhanced cases. 95.8% on non-contrast, with predefined failure handling for the rest.
Full validation in the paper Bland–Altman, contrast robustness, sex-based subgroup, failure taxonomy → Int. J. Comput. Assist. Radiol. Surg. · 2026 · DOI 10.1007/s11548-026-03606-2
08 The author
SF

Shih-Feng Huang, MD

Colorectal surgeon · KVGH · PhD candidate, NCKU

Robotic colorectal surgery practice at Kaohsiung Veterans General Hospital. Building the surgical-data-science layer that connects pre-operative CT, intra-operative video, and post-operative outcome — one open pipeline at a time.

Code Fork on GitHub Full pipeline. MIT-licensed. Reproduce these numbers on your own data. Paper Read the IJCARS paper Method, validation, limitations. The full surgical-data-science argument. Collaborate Email me Especially if you have a CRS cohort and want surgical-difficulty signal in your CT pipeline.
References

Full reference list

  1. [1] Heald RJ, Husband EM, Ryall RDH. The mesorectum in rectal cancer surgery—the clue to pelvic recurrence? Br J Surg. 1982;69(10):613.
  2. [2] de'Angelis N, Pigneur F, Martínez-Pérez A, et al. Predictors of surgical outcomes and survival in rectal cancer patients undergoing laparoscopic TME after neoadjuvant chemoradiation therapy. Oncotarget. 2018;9(38):25315.
  3. [3] Salerno G, Daniels IR, Brown G. Magnetic resonance imaging of the low rectum: defining the radiological anatomy. Colorectal Dis. 2006;8(Suppl 3):10.
  4. [4] Knol J, Keller DS. Total mesorectal excision technique—past, present, and future. Clin Colon Rectal Surg. 2020;33(3):134-143.
  5. [5] Yaşar NF, Gündoğdu E, Yılmaz AŞ, et al. Can 3D radiological calculations predict operational difficulties for rectal cancer? Medicine (Baltimore). 2024;103(3):e36961.
  6. [6] Escal L, Nougaret S, Guiu B, et al. MRI-based score to predict surgical difficulty in patients with rectal cancer. Br J Surg. 2018;105(1):140-146.
  7. [7] Killeen T, Banerjee S, Vijay V, et al. Magnetic resonance pelvimetry as a predictor of difficulty in laparoscopic operations for rectal cancer. Surg Endosc. 2010;24(12):2974-2979.
  8. [8] Iqbal A, Khan A, Hughes SJ, et al. Validation of a pelvic surgery difficulty risk model to predict difficult pelvic dissection and poor outcomes. Surgery. 2023;173(5):1199-1207.
  9. [9] Geitenbeek R, Baltus SC, Broekman M, et al. Multimodal machine learning for evaluating the predictive value of pelvimetric measurements for anastomotic leakage after restorative low anterior resection. Cancers (Basel). 2025;17(6):1051.
  10. [10] Ferko A, Malý O, Örhalmi J, Dolejš J. CT/MRI pelvimetry as a useful tool when selecting patients with rectal cancer for transanal total mesorectal excision. Surg Endosc. 2016;30(3):1164-1171.
  11. [11] Boyle KM, Chalmers AG, Finan PJ, Sagar PM, Burke D. Morphology of the mesorectum in patients with primary rectal cancer. Dis Colon Rectum. 2009;52(6):1122-1129.
  12. [12] Baltus SC, Geitenbeek RTJ, Frieben M, et al. Deep learning-based pelvimetry in pelvic MRI volumes for pre-operative difficulty assessment of total mesorectal excision. Surg Endosc. 2025;39(3):1536-1543.
  13. [13] Wasserthal J, Breit H, Meyer MT, et al. TotalSegmentator: robust segmentation of 104 anatomic structures in CT images. Radiol Artif Intell. 2023;5(5):e230024.
  14. [14] Chen W, Li Q, Fan Y, et al. Factors predicting difficulty of laparoscopic low anterior resection for rectal cancer with total mesorectal excision and double stapling technique. PLoS One. 2016;11(3):e0151773.
  15. [15] Baik SH, Kim NK, Lee KY, et al. Factors influencing pathologic results after total mesorectal excision for rectal cancer. Ann Surg Oncol. 2008;15(3):721-728.
  16. [16] Davenport MS, Perazella MA, Yee J, et al. Use of intravenous iodinated contrast media in patients with kidney disease. Radiology. 2020;294(3):660-668.
  17. [17] Gohmann RF, Gottschling S, Seitz P, et al. 3D segmentation and characterization of visceral and abdominal subcutaneous adipose tissue on CT: influence of contrast medium and contrast phase. Quant Imaging Med Surg. 2021;11(2):697-709.
  18. [18] Morsbach F, Zhang Y, Martin L, Lindqvist C, Brismar TB. Body composition evaluation with computed tomography: contrast media and slice thickness cause methodological errors. Nutrition. 2018;59:50-55.
  19. [19] Abitbol M. Evolution of the ischial spine and of the pelvic floor in the hominoidea. Am J Phys Anthropol. 1988;75(1):53-67.
  20. [20] Fischer B, Mitteroecker P. Allometry and sexual dimorphism in the human pelvis. Anat Rec (Hoboken). 2017;300(4):698-705.
  21. [21] Setiawati R, Rahardjo PP, Ruriana I, Guglielmi G. Anthropometric study using three-dimensional pelvic CT scan in sex determination among adult Indonesian population. Forensic Sci Med Pathol. 2023;19(1):24-31.
  22. [22] Korhonen UR, Solja R, Laitinen J, Heinonen S, Taipale P. MR pelvimetry measurements: analysis of inter- and intraobserver variation. Eur J Radiol. 2010;75(2):e56-e61.
  23. [23] Keller TM, Rake A, Michel S, et al. Obstetric MR pelvimetry: reference values and evaluation of inter- and intraobserver error and intraindividual variability. Radiology. 2003;227(1):37-43.