Machine learning‑based radiomics models accurately predict Crohn's disease‑related anorectal cancer
Boosting
Feature (linguistics)
Gradient boosting
DOI:
10.3892/ol.2024.14553
Publication Date:
2024-07-03T11:29:17Z
AUTHORS (15)
ABSTRACT
The radiological diagnosis of Crohn's disease (CD)‑related anorectal cancer is difficult; it often found in advanced stages and has a poor prognosis because the difficulty curative surgery. However, there are no studies on predicting CD‑related cancer. present study aimed to develop predictive model diagnose CD cancerous lesions more accurately way that can be interpreted by clinicians. Patients with who developed at Hyogo Medical University (Nishinomiya, Japan) between March 2009 June 2022 were included study. T2‑weighted T1‑weighted magnetic resonance (MR) images utilized for our analysis. Images segmented using open‑source 3D Slicer software, radiomic features extracted PyRadiomics. Six machine learning models investigated compared: i) Support vector machine; ii) naive Bayes; iii) random forest; iv) light gradient boosting v) extremely randomized trees; vi) regularized greedy forest (RGF). SHapley Additive exPlanations (SHAP) values calculated assess extent which each feature contributed model's predictions compared baseline, represented as average all test data. 28 patients 40 non‑cancer analyzed contrast‑enhanced 22 patients. highest area under curve (AUC) was RGF‑based constructed image features, achieving an AUC 0.944 (accuracy, 0.862; recall, 0.830). SHAP‑based explanation suggested strong association such complex lesion texture; greater pixel separation within same coronal cross‑section; larger, randomly distributed clumps pixels signal intensity; spherical shape images. MRI radiomics‑based RGF demonstrated outstanding performance These results may affect surveillance strategies colorectal
SUPPLEMENTAL MATERIAL
Coming soon ....
REFERENCES (39)
CITATIONS (0)
EXTERNAL LINKS
PlumX Metrics
RECOMMENDATIONS
FAIR ASSESSMENT
Coming soon ....
JUPYTER LAB
Coming soon ....