Whole-body uptake classification and prostate cancer staging in 68Ga-PSMA-11 PET/CT using dual-tracer learning.

PURPOSE: In PSMA-ligand PET/CT imaging, standardized evaluation frameworks and image-derived parameters are increasingly used to support prostate cancer staging. Clinical applicability remains challenging wherever manual measurements of numerous suspected lesions are required. Deep learning methods are promising for automated image analysis, typically requiring extensive expert-annotated image datasets to reach sufficient accuracy. We developed a deep learning method to support image-based staging, investigating the use of training information from two radiotracers. METHODS: In 173 subjects imaged with 68Ga-PSMA-11 PET/CT, divided into development (121) and test (52) sets, we trained and evaluated a convolutional neural network to both classify sites of elevated tracer uptake as nonsuspicious or suspicious for cancer and assign them an anatomical location. We evaluated training strategies to leverage information from a larger dataset of 18F-FDG PET/CT images and expert annotations, including transfer learning and combined training encoding the tracer type as input to the network. We assessed the agreement between the N and M stage assigned based on the network annotations and expert annotations, according to the PROMISE miTNM framework. RESULTS: In the development set, including 18F-FDG training data improved classification performance in four-fold cross validation. In the test set, compared to expert assessment, training with 18F-FDG data and the development set yielded 80.4% average precision [confidence interval (CI): 71.1-87.8] for identification of suspicious uptake sites, 77% (CI: 70.0-83.4) accuracy for anatomical location classification of suspicious findings, 81% agreement for identification of regional lymph node involvement, and 77% agreement for identification of metastatic stage. CONCLUSION: The evaluated algorithm showed good agreement with expert assessment for identification and anatomical location classification of suspicious uptake sites in whole-body 68Ga-PSMA-11 PET/CT. With restricted PSMA-ligand data available, the use of training examples from a different radiotracer improved performance. The investigated methods are promising for enabling efficient assessment of cancer stage and tumor burden.

18F-FDG PET/CT Uptake Classification in Lymphoma and Lung Cancer by Using Deep Convolutional Neural Networks.

Background Fluorine 18 (18F)-fluorodeoxyglucose (FDG) PET/CT is a routine tool for staging patients with lymphoma and lung cancer. Purpose To evaluate configurations of deep convolutional neural networks (CNNs) to localize and classify uptake patterns of whole-body 18F-FDG PET/CT images in patients with lung cancer and lymphoma. Materials and Methods This was a retrospective analysis of consecutive patients with lung cancer or lymphoma referred to a single center from August 2011 to August 2013. Two nuclear medicine experts manually delineated foci with increased 18F-FDG uptake, specified the anatomic location, and classified these findings as suspicious for tumor or metastasis or nonsuspicious. By using these expert readings as the reference standard, a CNN was developed to detect foci positive for 18F-FDG uptake, predict the anatomic location, and determine the expert classification. Examinations were divided into independent training (60%), validation (20%), and test (20%) subsets. Results This study included 629 patients (mean age, 52.2 years ± 20.4 [standard deviation]; 394 men). There were 302 patients with lung cancer and 327 patients with lymphoma. For the test set (123 patients; 10 782 foci), the CNN areas under the receiver operating characteristic curve (AUCs) for determining hypermetabolic 18F-FDG PET/CT foci that were suspicious for cancer versus nonsuspicious by using the five input features were as follows: CT alone, 0.78 (95% confidence interval [CI]: 0.72, 0.83); 18F-FDG PET alone, 0.97 (95% CI: 0.97, 0.98); 18F-FDG PET/CT, 0.98 (95% CI: 0.97, 0.99); 18F-FDG PET/CT maximum intensity projection (MIP), 0.98 (95% CI: 0.98, 0.99); and 18F-FDG PET/CT MIP atlas, 0.99 (95% CI: 0.98, 1.00). The combination of 18F-FDG PET and CT information improved overall classification accuracy (AUC, 0.975 vs 0.981, respectively; P < .001). Anatomic localization accuracy of the CNN was 2543 of 2639 (96.4%; 95% CI: 95.5%, 97.1%) for body part, 2292 of 2639 (86.9%; 95% CI: 85.3%, 88.5%) for region (ie, organ), and 2149 of 2639 (81.4%; 95% CI: 79.3%-83.5%) for subregion. Conclusion The fully automated anatomic localization and classification of fluorine 18-fluorodeoxyglucose PET uptake patterns in foci suspicious and nonsuspicious for cancer in patients with lung cancer and lymphoma by using a convolutional neural network is feasible and achieves high diagnostic performance when both CT and PET images are used. © RSNA, 2019 Online supplemental material is available for this article. See also the editorial by Froelich and Salavati in this issue.

Evaluation of an Automatic Classification Algorithm Using Convolutional Neural Networks in Oncological Positron Emission Tomography.

Introduction: Our aim was to evaluate the performance in clinical research and in clinical routine of a research prototype, called positron emission tomography (PET) Assisted Reporting System (PARS) (Siemens Healthineers) and based on a convolutional neural network (CNN), which is designed to detect suspected cancer sites in fluorine-18 fluorodeoxyglucose (18F-FDG) PET/computed tomography (CT). Method: We retrospectively studied two cohorts of patients. The first cohort consisted of research-based patients who underwent PET scans as part of the initial workup for diffuse large B-cell lymphoma (DLBCL). The second cohort consisted of patients who underwent PET scans as part of the evaluation of miscellaneous cancers in clinical routine. In both cohorts, we assessed the correlation between manually and automatically segmented total metabolic tumor volumes (TMTVs), and the overlap between both segmentations (Dice score). For the research cohort, we also compared the prognostic value for progression-free survival (PFS) and overall survival (OS) of manually and automatically obtained TMTVs. Results: For the first cohort (research cohort), data from 119 patients were retrospectively analyzed. The median Dice score between automatic and manual segmentations was 0.65. The intraclass correlation coefficient between automatically and manually obtained TMTVs was 0.68. Both TMTV results were predictive of PFS (hazard ratio: 2.1 and 3.3 for automatically based and manually based TMTVs, respectively) and OS (hazard ratio: 2.4 and 3.1 for automatically based and manually based TMTVs, respectively). For the second cohort (routine cohort), data from 430 patients were retrospectively analyzed. The median Dice score between automatic and manual segmentations was 0.48. The intraclass correlation coefficient between automatically and manually obtained TMTVs was 0.61. Conclusion: The TMTVs determined for the research cohort remain predictive of total and PFS for DLBCL. However, the segmentations and TMTVs determined automatically by the algorithm need to be verified and, sometimes, corrected to be similar to the manual segmentation.

Deep-Learning 18F-FDG Uptake Classification Enables Total Metabolic Tumor Volume Estimation in Diffuse Large B-Cell Lymphoma.

Total metabolic tumor volume (TMTV), calculated from 18F-FDG PET/CT baseline studies, is a prognostic factor in diffuse large B-cell lymphoma (DLBCL) whose measurement requires the segmentation of all malignant foci throughout the body. No consensus currently exists regarding the most accurate approach for such segmentation. Further, all methods still require extensive manual input from an experienced reader. We examined whether an artificial intelligence-based method could estimate TMTV with a comparable prognostic value to TMTV measured by experts. Methods: Baseline 18F-FDG PET/CT scans of 301 DLBCL patients from the REMARC trial (NCT01122472) were retrospectively analyzed using a prototype software (PET Assisted Reporting System [PARS]). An automated whole-body high-uptake segmentation algorithm identified all 3-dimensional regions of interest (ROIs) with increased tracer uptake. The resulting ROIs were processed using a convolutional neural network trained on an independent cohort and classified as nonsuspicious or suspicious uptake. The PARS-based TMTV (TMTVPARS) was estimated as the sum of the volumes of ROIs classified as suspicious uptake. The reference TMTV (TMTVREF) was measured by 2 experienced readers using independent semiautomatic software. The TMTVPARS was compared with the TMTVREF in terms of prognostic value for progression-free survival (PFS) and overall survival (OS). Results: TMTVPARS was significantly correlated with the TMTVREF (ρ = 0.76; P < 0.001). Using PARS, an average of 24 regions per subject with increased tracer uptake was identified, and an average of 20 regions per subject was correctly identified as nonsuspicious or suspicious, yielding 85% classification accuracy, 80% sensitivity, and 88% specificity, compared with the TMTVREF region. Both TMTV results were predictive of PFS (hazard ratio, 2.3 and 2.6 for TMTVPARS and TMTVREF, respectively; P < 0.001) and OS (hazard ratio, 2.8 and 3.7 for TMTVPARS and TMTVREF, respectively; P < 0.001). Conclusion: TMTVPARS was consistent with that obtained by experts and displayed a significant prognostic value for PFS and OS in DLBCL patients. Classification of high-uptake regions using deep learning for rapidly discarding physiologic uptake may considerably simplify TMTV estimation, reduce observer variability, and facilitate the use of TMTV as a predictive factor in DLBCL patients.