2026-10-15T04:05:08Z https://www.scidb.cn/oai

10.57760/sciencedb.27733 2026-03-12T10:35:02Z

2026-03-12 Automated Segmentation of Esophagus and Esophageal Squamous Cell Carcinoma on Non-Contrast CT: Development and Multi-Center Validation of a Dual-Layer CT-Derived Virtual Non-Contrast-Trained nnU-Net Model doi:10.57760/sciencedb.27733 en Objective: Accurate segmentation of the esophagus and esophageal squamous cell carcinoma (ESCC), especially early-stage lesions, on true non-contrast (TNC) CT is limited by poor soft-tissue contrast, while contrast-enhanced CT carries iodinated contrast-related risks. This study aimed to develop a virtual non-contrast (VNC)-trained nnU-Net model for automated TNC CT segmentation of the esophagus and ESCC, with stratified validation and clinical translation via lightweight software.Methods: 2D and 3D full-resolution nnU-Net models were trained on dual-layer CT-derived iodine map-guided VNC images from 200 cases (100 pathologically confirmed ESCC, 100 controls) at Center A. Models were validated on an 80-case independent test set (40 ESCC) and a 90-case external multi-center test set (45 ESCC), with pre-specified subgroup analysis for T1/T2 early-stage ESCC (n=144). Performance was evaluated via Dice Similarity Coefficient (DSC) and radiologists’ diagnostic AUC with/without model assistance. The optimal model was packaged into a standalone, Python-free, open-source inference software package.Results: The 3D model significantly outperformed the 2D model in the independent test set (esophagus DSC: 0.904 vs 0.866; ESCC DSC: 0.937 vs 0.843; all P<0.001) and external test set (esophagus DSC: 0.903 vs 0.850; ESCC DSC: 0.959 vs 0.819; all P<0.001). Model assistance significantly improved junior and senior radiologists’ AUC in both cohorts (all P<0.05). In the T1/T2 subgroup, the 3D model achieved an AUC of 0.932; model assistance lifted junior radiologists’ AUC from 0.750 to 0.910 (P<0.001) and senior radiologists’ AUC from 0.819 to 0.896 (P=0.010). The final inference software is open-source and runs independently without Python environment.Conclusion: The VNC-trained nnU-Net model enables accurate, robust esophagus and ESCC segmentation on routine non-contrast CT, especially for early-stage lesions. Combined with the lightweight open-source software, it holds substantial clinical application potential.Limitations: This study has several limitations. First, model training using a retrospective limited-sample dataset may compromise the model’s generalizability, warranting larger multi-center training cohorts. Second, the model was exclusively validated for ESCC, with untested performance for esophageal adenocarcinoma. Third, the 5 mm slice thickness of included CT images may increase the risk of missed detection for small z-axis lesions. Further validation in broader esophageal lesion cohorts, model optimization with thin-slice CT, and prospective real-world evaluation are required to confirm its broad applicability.Train and validate source code：Brief code for training a CT-based deep learning segmentation model of esophagus and esophageal squamous cell carcinoma using the nnU-Net framework. For more information on running inference with nnU-Net, click here: (https://github.com/MIC-DKFZ/nnUNet)。#0: activate virtual environmentconda create -n nnUNet python=3.10 -yconda activate nnUNetpip install nnunetv2pip install medpy#1: set environment variablesset nnUNet_raw_data_base="path\to\nnUNet_raw_data_base"set nnUNet_preprocessed="path\to\nnUNet_preprocessed"set RESULTS_FOLDER="path\to\nnUNet_trained_models"#2: Task11 (CT), Divide the data set"path\\Scripts\\nnUNetv2_convert_MSD_dataset.exe" -i "D:\nnunet\nnUNet\nnUNet_raw\Task11_example"#3: Preprocessing"path\\Scripts\\nnUNetv2_plan_and_preprocess.exe" -d 11 --verify_dataset_integrity#4: Training 2D models with 5 fold cross-validation"path\\Scripts\\nnUNetv2_train.exe" 11 2d all -tr nnUNetTrainer -device cudafor %%i in (0,1,2,3,4) do (echo Training fold %%icall "path\Scripts\nnUNetv2_train.exe" 11 2d %%i -tr nnUNetTrainer -device cuda)#5: Training 3D fullres models with 5 fold cross-validation"path\\Scripts\\nnUNetv2_train.exe" 11 3d_fullres all -tr nnUNetTrainer -device cudafor %%i in (0,1,2,3,4) do (echo Training fold %%icall "path\Scripts\nnUNetv2_train.exe" 11 3d_fullres %%i -tr nnUNetTrainer -device cuda)#6: Test set reasoning 2D and 3D fullres"path\\Scripts\\nnUNetv2_predict.exe" -i D:\nnunet\nnUNet\nnUNet_raw\Dataset011_example\imagesTs -o output -d 11 -c 3d_fullres -f all"path\\Scripts\\nnUNetv2_predict.exe" -i D:\nnunet\nnUNet\nnUNet_raw\Dataset011_example\imagesTs -o output -d 11 -c 2d -f all#7: Compute the test set DSC"path\\Scripts\\nnUNetv2_evaluate_folder.exe" -djfile output\dataset.json -pfile output\plans.json D:\nnunet\nnUNet\nnUNet_raw\Task11_example\labelsTs output#8: Computing Test Set HD95"path\\python.exe" calculate_hd95_multilabel.py --pred_folder output --gt_folder D:\nnunet\nnUNet\nnUNet_raw\Task11_example\labelsTs --labels 1,2 --output hd95_multilabel_results.json esophageal cancer; automatic segmentation; terminal software; deep learning Hao Wang PUBLIC https://creativecommons.org/licenses/by-nc/4.0/ dataset Science Data Bank