Ensemble deep learning model based on CT scans: differentiating and subtype-classifying pancreatic inflammations and tumors, and predicting pancreatic lesion invasiveness

Sep 23, 2026

Uploaded by Mabel

Pancreatic diseases, including pancreatitis and tumors, remain challenging to distinguish on imaging because of overlapping morphological features. This multi-center study develops an ensemble deep learning model using CT images to differentiate and subtype-classify pancreatic inflammations and tumors, and to predict lesion invasiveness. Using segmentation and diagnostic modules with Grad-CAM–based reference follow-up, the model is evaluated with Dice, IoU, sensitivity, and accuracy.

LanguageEnglish

CategoryHealthcare

FormatPDF

Pages15

Reading Time38 min

AccessFree to download

Background

Objective

Methods

Model development and evaluation

Invasiveness prediction

Results

Conclusions

Introduction

Original Article Ensemble deep learning model based on CT scans: differentiating and subtype-classifying pancreatic inflammations and tumors, and predicting pancreatic lesion invasiveness Xuhang Pan1\#, Qian Yang2\#, Maofen Shi1, Yupeng He1, Kun Qin1, Jinchao Zhu3, Tianle Zhang4, Hao Wu5, Rui Du6, Min Sun7, Suping Chen8, Hongyi Yang1, Yuhui Fang1, Suming Zhang1, Bo Yang1 1Institute of Medical Imaging, Department of Radiology, Taihe Hospital, Hubei University of Medicine, Shiyan, China; 2Department of Radiology, Hubei Cancer Hospital, Tongji Medical College, Huazhong University of Science and Technology, Wuhan, China; 3Department of Pathology, The Ninth People’s Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China; 4Institute for Network Sciences and Cyberspace, Tsinghua University, Beijing, China; 5Department of Radiology, Mayo Clinic, Rochester, MN, USA; 6Department of Radiology, Wuhan Children's Hospital, Tongji Medical College, Huazhong University of Science & Technology, Wuhan, China; 7Department of General Surgery, Taihe Hospital, Hubei University of Medicine, Shiyan, China; 8Advanced Application Team, GE Healthcare, Shanghai, China Contributions: (I) Conception and design: B Yang, S Zhang; (II) Administrative support: B Yang; (III) Provision of study materials or patients: B Yang, Q Yang, J Zhu; (IV) Collection and assembly of data: X Pan, Q Yang, M Shi, R Du, M Sun, S Chen, H Yang, Y Fang; (V) Data analysis and interpretation: X Pan, Q Yang; (VI) Manuscript writing: All authors; (VII) Final approval of manuscript: All authors. \#These authors contributed equally to this work. Correspondence to: Bo Yang, PhD; Suming Zhang, MD. Institute of Medical Imaging, Department of Radiology, Taihe Hospital, Hubei University of Medicine, 32 South Renmin Road, Maojian District, Shiyan 442000, [China. Email: yangbo1108@whu.edu.cn](China. Email: yangbo1108@whu.edu.cn); [304607346@qq.com](304607346@qq.com). Background: Pancreatic diseases, including pancreatitis and tumors, are often difficult to distinguish on imaging due to overlapping morphological features. The accurate classification of pancreatic lesions and preoperative prediction of invasiveness are critical for clinical decision-making and prognostic evaluation. The objective of this study was to develop an ensemble deep learning (DL) model based on computed tomography (CT) images for differentiating and subtyping pancreatic inflammations and tumors, as well as for predicting lesion invasiveness. Methods: This multi-center research included 6,740 patients’ pancreatic CT images. An ensemble DL model integrating DeepLabV3, nnUNet-MS, and Adaptive Pyramidal Shifted Window-SwinTransformer, respectively responsible for pancreatic segmentation, lesion segmentation, and diagnosis, was developed. Segmentation performance was evaluated using Dice and Intersection over Union (IoU); lesion differentiating and sub-classifying performance were assessed using sensitivity and accuracy. Gradientweighted Class Activation Mapping (Grad-CAM) was conducted for conservatively treated pancreatic ductal adenocarcinoma (PDAC) to predict 12-month invasiveness, with imaging follow-up as reference standard. Results: The DeepLabV3 module had good pancreas segmentation performance (internal test set: Dice: 0.983; IoU: 0.971; external test set Dice: 0.981; IoU: 0.969) . The nnUNet-MS module demonstrated superior lesion segmentation performance (internal validation set: Dice 0.941, IoU 0.932; external test set: Dice 0.942, IoU 0.930) to five other open-source nnUNet algorithms. The ensemble DL model showed high accuracy in both differentiating inflammatory and tumor lesions (internally 95.1%, externally 95.8%), as well as sub-classifying five inflammation subtypes and six tumor subtypes (internally 88.0% and externally 87.5%) . It exhibited high sensitivity for detecting PDAC [93.9%(95% CI: 91.0–96.0%) internally and 92.9%(95% CI: 90.1–95.1%) externally] . I

FAQs

What clinical problem does the study address?

Distinguishing pancreatic inflammations from tumors on imaging and predicting pancreatic lesion invasiveness are often difficult due to overlapping features and the need for timely, accurate assessment.

How was the ensemble deep learning model constructed?

The ensemble integrates DeepLabV3 for pancreatic segmentation, nnUNet-MS for lesion segmentation, and an Adaptive Pyramidal Shifted Window-SwinTransformer for diagnosis, built from multi-center CT data.

Which metrics were used to evaluate performance?

Segmentation performance was assessed using Dice and Intersection over Union (IoU), while lesion differentiating and sub-classifying performance used sensitivity and accuracy.

How was 12-month invasiveness prediction performed?

Grad-CAM was conducted for conservatively treated pancreatic ductal adenocarcinoma (PDAC), and imaging follow-up served as the reference standard for 12-month invasiveness.

Sep 23, 2026

Uploaded by Mabel