Enhancing Radiologist Performance in PTC Detection on Cervical Non-Contrast CT via a Deployment-Free Edge-Side 3D nnU-Net Software: A Multicenter Study

Background: Thyroid ultrasound accuracy depends heavily on operator expertise. This study developed a 3D deep learning model using non-contrast CT (NCCT) for joint segmentation of the thyroid gland and papillary thyroid carcinoma (PTC) and evaluated its clinical utility across radiologists with varying experience levels.Materials and Methods: This multicenter study included three cohorts: a training set: Center A, n=300, 200 patients with PTC, 100 normal control subjects (NCS), a temporal test set: Center A, n=120, 60 patients with PTC, 30 benign thyroid nodules (BTN), 30 NCS, and an external test set (Center B, n=100, 50 patients with PTC, 25 BTN, 25 NCS. 3D full-resolution and 2D models were developed using the nnU-Net framework and evaluated via five-fold cross-validation. The total segmentation model was also used for comparing thyroid segmentation. Performance was measured by Dice similarity coefficient (DSC), Recall and 95% Hausdorff distance (HD95). Two radiologists independently diagnosed PTC on NCCT with and without AI assistance at thyroid lobe level, and compared the model with the diagnostic performance of ultrasound TI-RADS. The top-performing model was deployed as a standalone, GUI-based Windows software for clinical application.Results: For PTC nodules segmentation, the 3D model significantly outperformed the 2D model across two test sets (DSC: 0.773/0.719 vs. 0.598/0.442; HD95: 2.00/2.24 mm vs. 3.17/5.66 mm; P<0.001). For thyroid gland segmentation, the 3D model also significantly outperformed the 2D model and total segmentation model (P<0.001). AI assistance significantly improved the junior radiologist’s AUC (0.765→0.888 in temporal and 0.717→0.859 in external sets; P<0.001), the senior radiologist’s AUC also improved (0.877→0.895, P=0.277; 0.832→0.881, P=0.010). Reading times were reduced by >40% for both radiologists. The developed software could complete conversion, inferencing, and rendering with one click, and can be seamlessly integrated into the diagnosis process.Conclusion: We developed a high-performance AI software for simultaneous thyroid gland and PTC nodule segmentation on cervical NCCT. This software significantly enhances diagnostic accuracy, especially for less-experienced radiologists, and provides a practical, contrast-free approach for preoperative PTC evaluation with excellent clinical translation potential.Model training and source code: Brief code for training a CT-based deep learning segmentation model of thyroid and Papillary thyroid 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