Canonical text (the exact bytes fingerprinted)

TRUVACE RECORD VERSION
record: TRV-2026-1913
version: 1
kind: certified
reason: Certified into the record
timestamp: 2026-10-19T18:12:52.931982Z
status: published
lens: trace
sector: health
headline: Habitat Radiomics and Machine Learning for Preoperative Three-class Classification of Pediatric Acute Appendicitis: A Dual-center Study with External Validation
dek: Accurate preoperative differentiation of pediatric acute appendicitis subtypes is relevant to treatment planning. We developed and externally validated a machine-learning framework combining habitat-based computed tomography (CT) radiomics with clinical variables for three-class classification (simple, complex, and appendiceal abscess). This retrospective dual-center study included 317 children undergoing appendectomy at Institution A (training n = 222, internal validation n = 95) and 174 children at Institution…
gain_title: A machine-learning model combining habitat-based CT radiomics and clinical variables distinguished simple, complex, and abscess appendicitis in children at a second hospital with a macro-AUC of 0.946.
problem_title: The model's probability outputs were poorly calibrated (slopes 0.23-0.30) and need recalibration before probabilistic use, and it is only a stratification aid, not a stand-alone diagnostic tool.
trace_subject: Preoperative CT-based three-class classification of pediatric acute appendicitis
gain_reading: A machine-learning model combining habitat-based CT radiomics and clinical variables distinguished simple, complex, and abscess appendicitis in children at a second hospital with a macro-AUC of 0.946.
gain_evidence: naive Bayes achieved the highest macro-AUC (0.946, 95% CI 0.910-0.974) | Habitat radiomics combined with machine learning provided accurate, externally validated three-class classification
problem_reading: The model's probability outputs were poorly calibrated (slopes 0.23-0.30) and need recalibration before probabilistic use, and it is only a stratification aid, not a stand-alone diagnostic tool.
problem_evidence: Calibration slopes (0.23-0.30) indicated that probability outputs require recalibration before probabilistic use | The model is a subtype-stratification aid rather than a stand-alone diagnostic or treatment-selection tool.
quick_read: Researchers built a classifier that uses preoperative noncontrast CT scans and clinical variables to sort pediatric appendicitis into simple, complex, and abscess subtypes. The CT volumes were divided into three habitat sub-regions, from which 1218 radiomic features were extracted, and ten selected features were combined with 20 clinical variables. Eight classifiers were compared, with naive Bayes performing best.

Earlier knowledge of the subtype could help treatment planning, but the tool is positioned only as a stratification aid. Its probability estimates were not well calibrated, the data were retrospective, and all children had undergone appendectomy, so performance in other settings or clinical workflows remains untested.
limitation: Probability outputs were poorly calibrated and require recalibration, and the model is described only as a subtype-stratification aid, not a stand-alone diagnostic or treatment-selection tool.
tag: Dual reading
key_points: The study was retrospective and dual-center: 317 children at Institution A for training and internal validation, and 174 children at Institution B for external validation. | Naive Bayes reached external accuracy 0.776 and macro-F1 0.766, significantly exceeding all seven comparator classifiers. | Per-class external AUCs were 0.920 for simple, 0.948 for complex, and 0.971 for abscess appendicitis. | In the ablation, fusing clinical variables with habitat radiomics gave the best external macro-AUC at 0.971, while conventional whole-VOI radiomics alone scored 0.582.
rundown: The cohort came from two institutions. Institution A supplied 317 children split into 222 for training and 95 for internal validation, and Institution B supplied 174 children as a fully external test set. All had undergone appendectomy and preoperative noncontrast CT.

Habitat segmentation used K-means++ with k = 3. Feature selection was done on the training cohort only, and the model comparison used bootstrap confidence intervals, DeLong tests, calibration statistics, and decision curve analysis.

A five-arm ablation showed habitat radiomics carried most of the signal. Habitat radiomics alone reached 0.952 external macro-AUC, compared with 0.763 for clinical variables alone and 0.582 for conventional whole-VOI radiomics alone. The fusion model reached 0.971.
sources:
- peer_reviewed | Journal of Imaging Informatics in Medicine | https://doi.org/10.1007/s10278-026-02383-3 | 2026-10-14
prev: 0000000000000000000000000000000000000000000000000000000000000000

sha256

615a6f1e9d50a4f4e841c88528cc80096dc54a79f1bf4d7bcd320428e2d289ea

previous

0000000000000000000000000000000000000000000000000000000000000000