Fully Automatic Quantitative Measurement of 18F-FDG PET/CT in Thymic Epithelial Tumors Using a Convolutional Neural Network

Abstract

Objectives: The aim of this study was to develop a deep learning (DL)-based segmentation algorithm for automatic measurement of metabolic parameters of 18 F-FDG PET/CT in thymic epithelial tumors (TETs), comparable performance to manual volumes of interest. Patients and Methods: A total of 186 consecutive patients with resectable TETs and preoperative 18 F-FDG PET/CTwere retrospectively enrolled (145 thymomas, 41 thymic carcinomas). A quasi-3D U-net architecture was trained to resemble ground-truth volumes of interest. Segmentation performance was assessed using the Dice similarity coefficient. Agreements between manual and DL-based automated extraction of SUV max , metabolic tumor volume (MTV), total lesion glycolysis (TLG), and 63 radiomics features were evaluated via concordance correlation coefficients (CCCs) and linear regression slopes. Diagnostic and prognostic values were compared in terms of area under the receiver operating characteristics curve (AUC) for thymic carcinoma and hazards ratios (HRs) for freedom from recurrence. Results: The mean Dice similarity coefficient was 0.83 ± 0.34. Automatically measured SUV max (slope, 0.97; CCC, 0.92), MTV (slope, 0.94; CCC, 0.96), and TLG (slope, 0.96; CCC, 0.96) were in good agreement with manual measurements. The mean CCC and slopes were 0.88 ± 0.06 and 0.89 ± 0.05, respectively, for the radiomics parameters. Automatically measured SUV max , MTV, and TLG showed good diagnostic accuracy for thymic carcinoma (AUCs: SUV max , 0.95; MTV, 0.85; TLG, 0.87) and significant prognostic value (HRs: SUV max , 1.31 [95% confidence interval, 1.16-1.48]; MTV,; TLG,). No significant differences in the AUCs or HRs were found between automatic and manual measurements for any of the metabolic parameters. Conclusions: Our DL-based model provides comparable segmentation performance and metabolic parameter values to manual measurements in TETs.

Fully Automatic Quantitative Measurement of Equilibrium Radionuclide Angiocardiography Using a Convolutional Neural Network

“…A set of pROIs and 24-frame ERNA images were randomly allocated into training (75%) and test datasets (25%). To evaluate the performance of the proposed model and prevent overfitting, we performed 4-fold cross-validation, resulting in 4 lots of 25% test results 11,12 . All test sets were independent of the training sets.…”

“…To evaluate the performance of the proposed model and prevent overfitting, we performed 4-fold cross-validation, resulting in 4 lots of 25% test results. 11,12 All test sets were independent of the training sets. The overall estimate of the model's performance was obtained by merging all test results from the 4 test folds.…”

Deep learning-based measurement of split glomerular filtration rate with 99mTc-diethylenetriamine pentaacetic acid renal scan

“…A set of images (summed over 2–3 min with a resolution of 256 × 256 pixels) and their associated GT ROIs were randomly divided into training (75%) and testing (25%) groups. To assess the model’s efficacy and avoid the risk of overfitting, a four-fold cross-validation-like rotation was executed, yielding four independent test groups, each comprising 25% of the data, with each test group being distinct from the training data [ 11 , 23 ]. The model’s overall performance was determined by aggregating the results from all four test folds.…”

[18F]fluorodeoxyglucose positron emission tomography/computed tomography characteristics of primary mediastinal germ cell tumors

“…Recently, the role of [ 18 F]fluorodeoxyglucose positron emission tomography/computed tomography ([ 18 F]FDG PET/CT) in the diagnosis and prognostic stratification of mediastinal tumors has been investigated 8 , 9 . However, there are limited [ 18 F]FDG PET/CT-related studies on primary MGCTs, and those that have been published mainly consist of case reports and small series.…”