Skip to content
MOHS AI
Published 2026 · 408-patient Sydney cohort

Predict which Mohs cases will need ≥13 sectionsbefore the patient sits down.

An ensemble of 30 machine-learning models trained on 408 consecutive procedures identifies complex cases from pre-operative clinical features — improving scheduling, counselling, and resource allocation.

Try the full predictor See the evidence

Yagiz Alp Aksoy, Simon Lee, Gilberto Moreno-Bonilla · 2012–2017 · mohs.panacea-i.com

Live demo · runs in your browser<13 likely
P(≥13 sections)22.9%
Confidence: high
Tumour area (ellipse)1.41 cm²

Marker at 1.5 cm² — the manuscript's SHAP-derived clinical threshold.

0

Procedures analysed

0.000

Shipping CV AUC

0.0%

High-confidence accuracy

0

Algorithms evaluated

Pipeline

From twelve inputs to a calibrated decision.

The tool you just used in the hero is the shipping model. It runs a calibrated logistic regression in your browser in under a millisecond — no round-trip to a server, no model to wait on.

  1. Step 1

    Pre-operative input

    12 variables a clinician already has: age, sex, tumour size X and Y, location, histology subtype, recurrence status, biopsy method.

  2. Step 2

    Stacking ensemble

    30 ML algorithms evaluated; calibrated logistic regression ships for inference at <1 ms, ensemble serves static context.

  3. Step 3

    Calibrated probability

    Probability of ≥13 sections + confidence flag. Exceeds the 1.5 cm² SHAP threshold? Flag it for extended OR time.

Six lenses on the same study.

Every page traces back to a specific result in the manuscript. No fabricated statistics, no demo data.

Live ensemble

Predictor

Enter 12 pre-operative variables. Get the probability of ≥13 sections with a 95% confidence band and a high-confidence flag.

Open
30 models

Evidence

Table 1 cohort. Table 2 effect sizes. Table 3 leaderboard of all 30 algorithms evaluated in the manuscript.

Open
SHAP

Why

SHAP dependence plots reveal the 1.5 cm² tumour-area threshold that drives every high-section-count prediction.

Open
Counter-intuitive

H-zone paradox

H-zone is anatomically high-risk yet requires fewer sections — because L-zone tumours are ~6× larger. Interactive story.

Open
Operations

Clinical tools

Multi-room OR day scheduler, MBS revenue projector, defect-size estimator, and case-similarity finder.

Open
AI assistant

Ask MOHS AI

An AI assistant that answers patient-education questions and explains SHAP contributions in plain English.

Open
Feature importance

Tumour area dominates. Everything else is secondary.

SHAP analysis across the ensemble puts the ellipse-formula tumour area above every anatomical or surgeon factor — and the relationship has a clean threshold at ~1.5 cm².

Tumour Area (cm²)
0.141
Tumour Size X
0.086
Tumour Size Y
0.068
Aggressive Histopathology
0.046
Recurrence
0.035
Age
0.035
From the manuscript

“The stacking ensemble achieved the highest cross-validation AUC of 0.891 (95% CI 0.849–0.934) and test AUC of 0.884. Tumour area emerged as the strongest predictor (SHAP 0.141) … wide neural-network architectures outperformed deeper configurations.”

Aksoy YA, Lee S, Moreno-Bonilla G. Development and Validation of Machine Learning Models for Predicting 13 or More Sections in Mohs Micrographic Surgery. 2026.

Cohort

408 consecutive procedures from a high-volume Mohs surgical centre, Sydney. Mean age 68.5 ± 12.9. BCC 89.9%.

Outcome

Binary classification: ≥13 vs <13 sections. Cohort balance 47.8% / 52.2% — aligned with the MBS 31002 billing cut-off.

Validation

5-fold stratified CV on 80% train, held-out n=82 test. 70.7% of test cases are high-confidence → 91.4% accuracy.