To develop and validate deep learning (DL) segmentation models for accurate, MRI-less amyloid PET quantification through the Centiloid scale. In this retrospective study, two DL models (PET/CT and PET-only) were trained using 2774 18F-FBP PET/MRI scan pairs from the ADNI dataset, where synthetic CT were generated from T1w MRI and used as training data. External validation was performed on 424 PET/CT scans from the OASIS-3 dataset. The performance of the DL models was compared to the MRI-based FreeSurfer reference method and the template-based rPOP method. Statistical analyses included the calculation of mean absolute error (MAE), R-squared (R²), equivalence testing, and diagnostic accuracy for amyloid positivity (Centiloid > 20), with a focus on the 0–40 Centiloid ‘gray zone’. The PET/CT model demonstrated the highest accuracy, with a MAE of 5.79 Centiloid and a strong correlation (R² = 0.96). The PET-only and rPOP methods showed higher errors with MAEs of 10.8 and 14.45 Centiloid, respectively. The PET/CT model’s Centiloid values were shown to be equivalent to the MRI-derived values within a ± 5 margin, while the PET-only and rPOP methods were not. In the Centiloid gray zone subgroup (n = 93), the PET/CT model yielded the highest diagnostic accuracy of 90.3

INTRODUCTION:Although tau positron emission tomography (PET) imaging is effective for staging tau pathology, it is limited clinically by cost and availability. Machine learning models based on magnetic resonance imaging (MRI)- and amyloid PET-derived features may serve as useful screening tools for tau pathology. METHODS:Multiple machine learning models were developed to classify tau positivity in the Braak III/IV region using structural MRI, amyloid PET, and demographic features. Alzheimer's Disease Neuroimaging Initiative (ADNI) (n = 410) data were used for model training. Open Access Series of Imaging Studies (OASIS-3; n = 143) and the Standardized Centralized Alzheimer's Disease Neuroimaging (SCAN; n = 154) data were used for external validation. RESULTS:Logistic regression achieved the best performance with areas under the curve (AUCs) of 0.92 for both internal and external validation. Combined external validation yielded accuracy/sensitivity/specificity of 85%/83%/85%. Subjects with mild cognitive impairment and predicted tau positivity progressed to AD at a significantly faster pace (p < 10-6). DISCUSSION:Our model demonstrates the feasibility of classifying tau burden in amyloid-positive cohorts with MRI- and amyloid PET-derived features and may serve as a surrogate biomarker.

Lumbar spinal stenosis (LSS), prevalent among the elderly, involves the narrowing of the spinal canal or neural foramen, often causing nerve compression. Understanding the biomechanical effects of postural changes on the ligamentum flavum (LF) and other lumbar structures is crucial for LSS management. This study aimed to investigate the impact of different postures on LF thickness, disc height, and foraminal height to enhance our understanding of LSS pathogenesis. Nine cadaveric lumbar specimens were analyzed for biomechanical changes in LF thickness, intervertebral discs, and foraminal height across flexion, extension, lateral bending, and axial rotation postures. Postural changes were induced by applying joint moments, and specimens were imaged using computed tomography. Three-dimensional models were created to measure the spinal structures during different postures. LF thickness increased progressively from the upper to the lower lumbar segments, peaking at L4/L5. Flexion significantly reduced LF thickness at L3/L4 and L4/L5 levels, while lateral bending thickened LF at L1/L2 and thinned it at L4/L5, with no significant difference at the concave and convex sides. Right axial rotation increased LF thickness at L1/L2 and L2/L3. Disc height increased anteriorly during extension and decreased during flexion. Foraminal height decreased during extension at L1/L2, L2/L3, and L3/L4 and increased during flexion at L2/L3, L3/L4, and L4/L5. Lateral bending caused foraminal height changes, with decreases on the concave side and increases on the convex side. This study revealed that postural variations affect LF thickness, disc height, and foraminal height, providing insights into LSS biomechanics. These findings may aid in developing improved management strategies for individuals with LSS.

Objectives: Predicting intensive care unit (ICU) admissions during pandemic outbreaks such as COVID-19 can assist clinicians in early intervention and the better allocation of medical resources. Artificial intelligence (AI) tools are promising for this task, but their development can be hindered by the limited availability of training data. This study aims to explore model development strategies in data-limited scenarios, specifically in detecting the need for ICU admission using chest X-rays of COVID-19 patients by leveraging transfer learning and data extension to improve model performance. Methods: We explored convolutional neural networks (CNNs) pre-trained on either natural images or chest X-rays, fine-tuning them on a relatively limited dataset (COVID-19-NY-SBU, n = 899) of lung-segmented X-ray images for ICU admission classification. To further address data scarcity, we introduced a dataset extension strategy that integrates an additional dataset (MIDRC-RICORD-1c, n = 417) with different but clinically relevant labels. Results: The TorchX-SBU-RSNA and ELIXR-SBU-RSNA models, leveraging X-ray-pre-trained models with our training data extension approach, enhanced ICU admission classification performance from a baseline AUC of 0.66 (56% sensitivity and 68% specificity) to AUCs of 0.77-0.78 (58-62% sensitivity and 78-80% specificity). The gradient-weighted class activation mapping (Grad-CAM) analysis demonstrated that the TorchX-SBU-RSNA model focused more precisely on the relevant lung regions and reduced the distractions from non-relevant areas compared to the natural image-pre-trained model without data expansion. Conclusions: This study demonstrates the benefits of medical image-specific pre-training and strategic dataset expansion in enhancing the model performance of imaging AI models. Moreover, this approach demonstrates the potential of using diverse but limited data sources to alleviate the limitations of model development for medical imaging AI. The developed AI models and training strategies may facilitate more effective and efficient patient management and resource allocation in future outbreaks of infectious respiratory diseases.

Simultaneous PET/MR imaging enables precise anatomical localization and PET quantification by reducing PET-to-MR misalignments. However, involuntary motion during scans may still cause misalignment and quantification imprecision. Current mutual information (MI)-based co-registration methods do not account for the tissue-specific uptake patterns of PET and therefore could result in suboptimal alignment. To address this, we proposed a novel image co-registration method, namely the tracer characteristic-based co-registration (TCBC) method, which takes advantage of specific PET uptake patterns within a selected anatomical region to improve the image alignment and PET quantification. TCBC was evaluated using simulation and in vivo 18F-Florbetapir PET/MR data from the OASIS-3 dataset. In simulations, TCBC demonstrated superior alignment accuracy with lower root mean square error and higher R-squared values compared to the conventional MI-based co-registration from FreeSurfer in recovering the simulated patient motion. In the retrospective human study, we evaluated the detectability of age-related amyloid burden in healthy controls under different co-registration methods as a demonstrative use case. TCBC significantly enhanced the detectability of age-related amyloid burden with stronger correlations across all five regions of evaluation, such as the medial orbitofrontal cortex (p < 0.001), precuneus (p = 0.004), and early amyloid-β composite (p = 0.002), compared to FSMC (p = 0.004, 0.007, and 0.006, respectively) and uncorrected (p = 0.378, 0.023, and 0.039, respectively) methods. Bootstrap analyses also confirmed TCBC’s robustness in smaller samples, yielding tighter confidence intervals and lower means of p-values, such as 0.032 (95

Abstract Background The sagittal imbalance (SI) of spine triggers compensatory mechanisms (CMs) of lower extremity (LE) to restore trunk balance. These CMs can cause long‐period stress on the femur and may possibly alter the femoral morphology. This cross‐sectional observational study aimed to answer the following questions: (a) Do SI subjects exhibit greater femoral bowing compared to subjects with sagittal balance? (b) Are there associations between femoral bowing and CMs of LE in SI subjects? Methods Subjects who underwent biplanar full body radiographs with the EOS imaging system between January 2016 and September 2021 were recruited. Sagittal parameters included T1‐pelvic angle (TPA), pelvic incidence (PI), pelvic tilt (PT), sacral slope, lumbar lordosis (LL), PI‐LL, and PT/PI ratio. LE parameters were femoral obliquity angle (FOA), knee flexion angle (KA), and ankle dorsiflexion angle. Femoral bowing was quantified as 3D radius of femoral curvature (RFC). Associations between 3D RFC and the radiographic parameters were analyzed. Results A total of 105 subjects were included, classified into balance group (TPA < 14°, n = 40), SI group (TPA ≥ 14° and KA <5°, n = 30), and SI with knee flexion group (TPA ≥ 14° and KA ≥ 5°, n = 35). 3D RFC was significantly lower in SI with knee flexion group compared to the other two groups (both p < 0.001). Stepwise linear regression showed that age, SI and knee flexion, femoral length (FL), FOA, and KA were independent predictors for 3D RFC. Conclusion Greater femoral bowing is observed in subjects with SI and knee flexion compared to the balanced population. CM parameters, including KA and FOA, are associated with 3D RFC. Further longitudinal study is needed to investigate the cause‐and‐effect relationship between SI, CMs of LE, and femoral bowing.

Background:Class I echocardiographic guidelines in primary mitral regurgitation (PMR) risks left ventricular ejection fraction (LVEF) < 50% after mitral valve surgery even with pre-surgical LVEF > 60%. There are no models predicting LVEF < 50% after surgery in the complex interplay of increased preload and facilitated ejection in PMR using cardiac magnetic resonance (CMR). Objective:Use regression and machine learning models to identify a combination of CMR LV remodeling and function parameters that predict LVEF < 50% after mitral valve surgery. Methods:CMR with tissue tagging was performed in 51 pre-surgery PMR patients (median CMR LVEF 64%), 49 asymptomatic (median CMR LVEF 63%), and age-matched controls (median CMR LVEF 64%). To predict post-surgery LVEF < 50%, least absolute shrinkage and selection operator (LASSO), random forest (RF), extreme gradient boosting (XGBoost), and support vector machine (SVM) were developed and validated in pre-surgery PMR patients. Recursive feature elimination and LASSO reduced the number of features and model complexity. Data was split and tested 100 times and models were evaluated via stratified cross validation to avoid overfitting. The final RF model was tested in asymptomatic PMR patients to predict post-surgical LVEF < 50% if they had gone to mitral valve surgery. Results:Thirteen pre-surgery PMR had LVEF < 50% after mitral valve surgery. In addition to LVEF (P = 0.005) and LVESD (P = 0.13), LV sphericity index (P = 0.047) and LV mid systolic circumferential strain rate (P = 0.024) were predictors of post-surgery LVEF < 50%. Using these four parameters, logistic regression achieved 77.92% classification accuracy while RF improved the accuracy to 86.17%. This final RF model was applied to asymptomatic PMR and predicted 14 (28.57%) out of 49 would have post-surgery LVEF < 50% if they had mitral valve surgery. Conclusions:These preliminary findings call for a longitudinal study to determine whether LV sphericity index and circumferential strain rate, or other combination of parameters, accurately predict post-surgical LVEF in PMR.

Accurate prediction of MCI-to-AD progression is an important yet challenging task. We introduce a new quantitative parameter: the atrophy-weighted standard uptake value ratio (awSUVR), defined as the PET SUVR divided by the hippocampal volume measured with MR, and evaluate whether it may provide better prediction of the MCI-to-AD progression. Materials and Methods: We used ADNI data to evaluate the prediction performances of the awSUVR against SUVR. 571, 363 and 252 18-F-Florbetaipir scans were selected based on criteria of conversion at the third, fifth and seventh year after the PET scans, respectively. Corresponding MR scans were segmented with Freesurfer and applied on PET for SUVR and awSUVR computation. We also searched for the optimal combination of target and reference regions. In addition to evaluating the overall prediction performances, we also evaluated the prediction for APOE4 carriers and non-carriers. For the scans with false predictions, we used 18-F-Flortaucipir scans to investigate the potential source of error. Results: awSUVR provides more accurate prediction than the SUVR in all three progression criteria. The 5-year prediction accuracy/sensitivity/specificity is 90/81/93% for awSUVR and 86/81/88% for SUV. awSUVR also yields good 3- and 7-year prediction accuracy/sensitivity/specificity of 91/57/96 and 92/89/93, respectively. APOE4 carriers generally are slightly more difficult to predict for the progression. False negative prediction is found to either due to a near-cutoff mis-classification or potentially non-AD dementia pathology. False positive prediction is mainly due to the slightly delayed progression than the expected progression time. Conclusion: We demonstrated with ADNI data that 18-F-Florbetapir SUVR weighted with hippocampus volume may provide good prediction power with over 90% accuracy in MCI-to-AD progression.

We found increased binding of the translocator protein radioligand [18F]DPA-714 in several brain regions in fibromyalgia patients compared with healthy controls.

An accurate model to assess the risk of disease progression for mild cognitive impairment (MCI) patients may be clinically useful for patient selection in disease modifying therapies to prevent or slow the progression to Alzheimer’s disease (AD) dementia. We aim to examine whether quantitative amyloid PET scans with 18F-Florbetapir and structural MRI, under a newly proposed feature of the atrophy-weighted standardized uptake value ratio (awSUVR), may provide an accurate prediction of the MCI-to-AD progression at the 3-, 5- and 7-year intervals. Methods: We used data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) to develop the prediction model and to evaluate the prediction performances. Retrospective data from 354 MCI patients recruited and acquired by ADNI were used to develop and evaluate the prediction model. Inclusion of eligible subjects was based on the diagnosis at follow-up visits after the amyloid PET scan and whether the subject remained as MCI or converted to AD during the specified time periods. awSUVR was defined as the PET SUVR divided by the hippocampal volume measured with MRI. Predictive performances were evaluated both on the scan- and subject-specific levels. Results: 571, 363 and 252 18F-Florbetapir scans were included based on the selection criteria of either stable MCI or MCI-to-AD conversion at the third, fifth and seventh year after the PET scans with conversion rates of 12%, 28% and 45%, respectively. awSUVR provided more accurate prediction than the SUVR in all three progression criteria. In the scan-based analysis, the 5-year prediction accuracy/sensitivity/specificity was 90/81/93% for awSUVR and 84/74/89% for SUV. awSUVR yielded 3- and 7-year prediction accuracy/sensitivity/specificity of 90/60/94% and 90/89/92%, respectively. False predictions were likely due to delayed progression, near-cutoff misclassification, or non-Alzheimer’s pathology. Conclusion: We demonstrated that 18F-Florbetapir SUVR weighted with hippocampal volume may provide good prediction power with 90% accuracy in MCI-to-AD progression events.

ABSTRACT Objective To assess the presence of brain and systemic inflammation in subjects newly diagnosed with Parkinson’s disease (PD). Background Evidence for a pathophysiologic role of inflammation in PD is growing. However, several key gaps remain as to the role of inflammation in PD, including the extent of immune activation at early stages, potential effects of PD treatments on inflammation, and whether pro-inflammatory signals are associated with clinical features and/or predict more rapid progression. Methods We enrolled subjects with de novo PD (n=58) and age-matched controls (n=62). Subjects underwent clinical assessments, including the Movement Disorder Society-United Parkinson’s Disease rating scale (MDS-UPDRS). Comprehensive cognitive assessment meeting MDS Level II criteria for mild cognitive impairment (MCI) testing was performed. Blood was obtained for flow cytometry and cytokine/chemokine analyses. Subjects underwent imaging with 18 F-DPA-714, a translocator protein 18kd (TSPO) ligand, and lumbar puncture if eligible and consented. Results Baseline demographics and medical history were comparable between groups. PD subjects showed significant differences in University of Pennsylvania Smell Identification Test, Schwab and England Activities of Daily Living, Scales for Outcomes in PD autonomic dysfunction, and MDS-UPDRS scores. Cognitive testing demonstrated significant differences in cognitive composite, executive function, and visuospatial domain scores at baseline. PET imaging showed increased 18 F-DPA-714 signal in PD subjects. 18 F-DPA-714 signal correlated with several cognitive measures and some chemokines. Conclusions 18 F-DPA-714 imaging demonstrated increased central inflammation in de novo PD subjects compared to controls. Longitudinal follow-up will be important to determine whether the presence of inflammation predicts cognitive decline.

In this study, we modified the previously proposed X2CT-GAN to build a 2Dto3D-GAN of the spine. This study also incorporated the radiologist's perspective in the adjustment of input signals to prove the feasibility of the automatic production of three-dimensional (3D) structures of the spine from simulated bi-planar two-dimensional (2D) X-ray images. Data from 1012 computed tomography (CT) studies of 984 patients were retrospectively collected. We tested this model under different dataset sizes (333, 666, and 1012) with different bone signal conditions to observe the training performance. A 10-fold cross-validation and five metrics-Dice similarity coefficient (DSC) value, Jaccard similarity coefficient (JSC), overlap volume (OV), and structural similarity index (SSIM)- were applied for model evaluation. The optimal mean values for DSC, JSC, OV, SSIM anteroposterior (AP), and SSIM Lateral (Lat) were 0.8192, 0.6984, 0.8624, 0.9261, and 0.9242, respectively. There was a significant improvement in the training performance under empirically enhanced bone signal conditions and with increasing training dataset sizes. These results demonstrate the potential of the clinical implantation of GAN for automatic production of 3D spine images from 2D images. This prototype model can serve as a foundation in future studies applying transfer learning for the development of advanced medical diagnostic techniques.

Measuring amyloid and predicting tau status using a single amyloid PET study would be valuable for assessing brain AD pathophysiology. We hypothesized that early-frame amyloid PET (efAP) correlates with the presence of tau pathology because the initial regional brain concentrations of radioactivity are determined primarily by blood flow, which is expected to be decreased in the setting of tau pathology. Methods: The study included 120 participants (63 amyloid-positive and 57 amyloid-negative) with dynamic 18F-florbetapir PET and static 18F-flortaucipir PET scans obtained within 6 mo of each other. These subjects were predominantly cognitively intact in both the amyloidpositive (63%) and the amyloid-negative (93%) groups. Parameters for efAP quantification were optimized for stratification of tau PET positivity, assessed by either a tauopathy score or Braak regions. The ability of efAP to stratify tau positivity was measured using receiveroperating-characteristic analysis of area under the curve (AUC). Pearson r and Spearman p were used for parametric and nonparametric comparisons between efAP and tau PET, respectively. Standardized net benefit was used to evaluate improvement in using efAP as an additional copredictor over hippocampal volume in predicting tau PET positivity. Results: Measuring efAP within the hippocampus and summing the first 3 min of brain activity after injection showed the strongest discriminative ability to stratify for tau positivity (AUC, 0.67-0.89 across tau PET Braak regions) in amyloid-positive individuals. Hippocampal efAP correlated significantly with a global tau PET tauopathy score in amyloid-positive participants (r = -0.57, P< 0.0001). Compared with hippocampal volume, hippocampal efAP showed a stronger association with tau PET Braak stage (p = -0.58 vs. -0.37) and superior stratification of tau PET tauopathy score (AUC, 0.86 vs. 0.66; P= 0.002). Conclusion: Hippocampal efAP can provide additional information to conventional amyloid PET, including estimation of the likelihood of tau positivity in amyloid-positive individuals.

This study aimed to build machine learning prediction models for predicting pathological subtypes of prevascular mediastinal tumors (PMTs). The candidate predictors were clinical variables and dynamic contrast–enhanced MRI (DCE-MRI)–derived perfusion parameters. The clinical data and preoperative DCE–MRI images of 62 PMT patients, including 17 patients with lymphoma, 31 with thymoma, and 14 with thymic carcinoma, were retrospectively analyzed. Six perfusion parameters were calculated as candidate predictors. Univariate receiver-operating-characteristic curve analysis was performed to evaluate the performance of the prediction models. A predictive model was built based on multi-class classification, which detected lymphoma, thymoma, and thymic carcinoma with sensitivity of 52.9%, 74.2%, and 92.8%, respectively. In addition, two predictive models were built based on binary classification for distinguishing Hodgkin from non-Hodgkin lymphoma and for distinguishing invasive from noninvasive thymoma, with sensitivity of 75% and 71.4%, respectively. In addition to two perfusion parameters (efflux rate constant from tissue extravascular extracellular space into the blood plasma, and extravascular extracellular space volume per unit volume of tissue), age and tumor volume were also essential parameters for predicting PMT subtypes. In conclusion, our machine learning–based predictive model, constructed with clinical data and perfusion parameters, may represent a useful tool for differential diagnosis of PMT subtypes.

There is a growing interest in using 18F-DPA-714 PET to study neuroinflammation and microglial activation through imaging the 18-kDa translocator protein (TSPO). Although quantification of 18F-DPA-714 binding can be achieved through kinetic modeling analysis with an arterial input function (AIF) measured with blood sampling procedures, the invasiveness of such procedures has been an obstacle for wide application. To address these challenges, we developed an image-derived input function (IDIF) that noninvasively estimates the arterial input function from the images acquired for 18F-DPA-714 quantification. Methods: The method entails three fully automatic steps to extract the IDIF, including a segmentation of voxels with highest likelihood of being the arterial blood over the carotid artery, a model-based matrix factorization to extract the arterial blood signal, and a scaling optimization procedure to scale the extracted arterial blood signal into the activity concentration unit. Two cohorts of human subjects were used to evaluate the extracted IDIF. In the first cohort of five subjects, arterial blood sampling was performed, and the calculated IDIF was validated against the measured AIF through the comparison of distribution volumes from AIF (VT,AIF) and IDIF (VT,IDIF). In the second cohort, PET studies from twenty-eight healthy controls without arterial blood sampling were used to compare VT,IDIF with VT,REF measured using a reference region-based analysis to evaluate whether it can distinguish high-affinity (HAB) and mixed-affinity (MAB) binders. Results: In the arterial blood-sampling cohort, VT derived from IDIF was found to be an accurate surrogate of the VT from AIF. The bias of VT, IDIF was −5.8 ± 7.8% when compared to VT,AIF, and the linear mixed effect model showed a high correlation between VT,AIF and VT, IDIF (p < 0.001). In the nonblood-sampling cohort, VT, IDIF showed a significance difference between the HAB and MAB healthy controls. VT, IDIF and standard uptake values (SUV) showed superior results in distinguishing HAB from MAB subjects than VT,REF. Conclusions: A novel IDIF method for 18F-DPA-714 PET quantification was developed and evaluated in this study. This IDIF provides a noninvasive alternative measurement of VT to quantify the TSPO binding of 18F-DPA-714 in the human brain through dynamic PET scans.

The aim of the study was to use a previously proposed mask region–based convolutional neural network (Mask R-CNN) for automatic abnormal liver density detection and segmentation based on hepatocellular carcinoma (HCC) computed tomography (CT) datasets from a radiological perspective. Training and testing datasets were acquired retrospectively from two hospitals of Taiwan. The training dataset contained 10,130 images of liver tumor densities of 11,258 regions of interest (ROIs). The positive testing dataset contained 1,833 images of liver tumor densities with 1,874 ROIs, and negative testing data comprised 20,283 images without abnormal densities in liver parenchyma. The Mask R-CNN was used to generate a medical model, and areas under the curve, true positive rates, false positive rates, and Dice coefficients were evaluated. For abnormal liver CT density detection, in each image, we identified the mean area under the curve, true positive rate, and false positive rate, which were 0.9490, 91.99%, and 13.68%, respectively. For segmentation ability, the highest mean Dice coefficient obtained was 0.8041. This study trained a Mask R-CNN on various HCC images to construct a medical model that serves as an auxiliary tool for alerting radiologists to abnormal CT density in liver scans; this model can simultaneously detect liver lesions and perform automatic instance segmentation.

Abstract Objectives The aim of this study was to develop and validate a prognostic model incorporating [18F]FDG PET/CT radiomics for patients of minor salivary gland carcinoma (MSGC). Methods We retrospectively reviewed the pretreatment [18F]FDG PET/CT images of 75 MSGC patients treated with curative intent. Using a 1.5:1 ratio, the patients were randomly divided into a training and validation group. The main outcome measurements were overall survival (OS) and relapse-free survival (RFS). All of the patients were followed up for at least 30 months or until death. Following segmentation of tumors and lymph nodes on PET images, radiomic features were extracted. The prognostic significance of PET radiomics and clinical parameters in the training group was examined using receiver operating characteristic curve analysis. Variables showing a significant impact on OS and RFS were entered into multivariable Cox regression models. Recursive partitioning analysis was subsequently implemented to devise a prognostic index, whose performance was examined in the validation group. Finally, the performance of the index was compared with clinical variables in the entire cohort and nomograms for surgically treated cases. Results The training and validation groups consisted of 45 and 30 patients, respectively. The median follow-up time in the entire cohort was 59.5 months. Eighteen relapse, 19 dead, and thirteen relapse, eight dead events were found in the training and validation cohorts, respectively. In the training group, two factors were identified as independently associated with poor OS, i.e., (1) tumors with both high maximum standardized uptake value (SUVmax) and discretized intensity entropy and (2) poor performance status or N2c-N3 stage. A prognostic model based on the above factors was devised and showed significant higher concordance index (C-index) for OS than those of AJCC stage and high-risk histology (C-index: 0.83 vs. 0.65, P = 0.005; 0.83 vs. 0.54, P < 0.001, respectively). This index also demonstrated superior performance than nomogram for OS (C-index: 0.88 vs. 0.70, P = 0.017) and that for RFS (C-index: 0.87 vs. 0.72, P = 0.004). Conclusions We devised a novel prognostic model that incorporates [18F]FDG PET/CT radiomics and may help refine outcome prediction in patients with MSGC.

Purpose Previous studies have shown that SUVmax on 18F-FDG PET/CT predicts prognosis in patients with salivary gland carcinoma (SGC). Here, we sought to evaluate whether texture features extracted from 18F-FDG PET/CT images may provide additional prognostic information for SGC with high-risk histology. Methods We retrospectively examined pretreatment 18F-FDG PET/CT images obtained from 85 patients with nonmetastatic SGC showing high-risk histology. All patients were treated with curative intent. We used the fixed threshold of 40% of SUVmax for tumor delineation. PET texture features were extracted by using histogram analysis, normalized gray-level co-occurrence matrix, and gray-level size zone matrix. Optimal cutoff points for each PET parameter were derived from receiver operating characteristic curve analyses. Recursive partitioning analysis was used to construct a prognostic model for overall survival (OS). Results Receiver operating characteristic curve analyses revealed that SUVmax, SUV entropy, uniformity, entropy, zone-size nonuniformity, and high-intensity zone emphasis were significantly associated with OS. The strongest associations with OS were found for high SUVmax (>6.67) and high SUV entropy (>2.50). Multivariable Cox analysis identified high SUVmax, high SUV entropy, performance status, and N2c–N3 stage as independent predictors of survival. A prognostic model derived from multivariable analysis revealed that patients with high SUVmax and SUV entropy or with the presence of poor performance status or N2c–N3 were associated with worse OS. Conclusions A prognostic model that includes SUVmax and SUV entropy is useful for risk stratification and supports the additional benefit of texture analysis for SGC with high-risk histology.

SummaryObjectiveDetection of focal cortical dysplasia (FCD) is of paramount importance in epilepsy presurgical evaluation. Our study aims at utilizing quantitative positron emission tomography (QPET) analysis to complement magnetic resonance imaging (MRI) postprocessing by a morphometric analysis program (MAP) to facilitate automated identification of subtle FCD.MethodsWe retrospectively included a consecutive cohort of surgical patients who had a negative preoperative MRI by radiology report. MAP was performed on T1‐weighted volumetric sequence and QPET was performed on PET/computed tomographic data, both with comparison to scanner‐specific normal databases. Concordance between MAP and QPET was assessed at a lobar level, and the significance of concordant QPET‐MAP+ abnormalities was confirmed by postresective seizure outcome and histopathology. QPET thresholds of standard deviations (SDs) of −1, −2, −3, and −4 were evaluated to identify the optimal threshold for QPET‐MAP analysis.ResultsA total of 104 patients were included. When QPET thresholds of SD = −1, −2, and −3 were used, complete resection of the QPET‐MAP+ region was significantly associated with seizure‐free outcome when compared with the partial resection group (P = 0.023, P < 0.001, P = 0.006) or the no resection group (P = 0.002, P < 0.001, P = 0.001). The SD threshold of −2 showed the best combination of positive rate (55%), sensitivity (0.68), specificity (0.88), positive predictive value (0.88), and negative predictive value (0.69). Surgical pathology of the resected QPET‐MAP+ areas revealed mainly FCD type I. Multiple QPET‐MAP+ regions were present in 12% of the patients at SD = −2.SignificanceOur study demonstrates a practical and effective approach to combine quantitative analyses of functional (QPET) and structural (MAP) imaging data to improve identification of subtle epileptic abnormalities. This approach can be readily adopted by epilepsy centers to improve postresective seizure outcomes for patients without apparent lesions on MRI.

The unidentified presence of uterine smooth muscle malignancies poses a tremendous risk in women planning surgery for presumed benign leiomyomas. We sought to investigate whether preoperative FDG PET may be useful to identify leiomyosarcomas (LMS) and smooth muscle tumors of uncertain malignant potential (STUMP). Methods . We investigated patients with rapidly growing uterine masses which were suspected of being malignant on ultrasound or MRI. Among the 21 patients who underwent FDG PET, we identified 7 LMS, 1 STUMP, and 13 leiomyomas. PET-derived parameters and FDG uptake patterns were analyzed retrospectively. Results . The SUV max values of LMS/STUMP (range: 3.7–11.8) were significantly higher than those observed in leiomyomas (range: 2.0–9.4; P=0.003) despite a significant overlap. The metabolic tumor/necrosis ratio was significantly higher in LMS/STUMP than in leiomyomas (P<0.001), with no significant intergroup overlaps. All LMS/STUMP revealed a characteristic pattern of FDG uptake, identifying a specific “hollow ball” sign (corresponding to areas of coagulative tumor necrosis). In contrast, this sign was invariably absent in patients with leiomyomas. Conclusion . The characteristic FDG uptake pattern instead of SUV on PET images allows identifying LMS/STUMP in patients with rapidly growing uterine masses, avoiding the deleterious consequences of regular surgery for presumed benign leiomyomas.