Manual prostate MRI segmentation by readers with different experience: a study of the learning progress.

OBJECTIVE: To evaluate the learning progress of less experienced readers in prostate MRI segmentation. MATERIALS AND METHODS: One hundred bi-parametric prostate MRI scans were retrospectively selected from the Göteborg Prostate Cancer Screening 2 Trial (single center). Nine readers with varying degrees of segmentation experience were involved: one expert radiologist, two experienced radiology residents, two inexperienced radiology residents, and four novices. The task was to segment the whole prostate gland. The expert's segmentations were used as reference. For all other readers except three novices, the 100 MRI scans were divided into five rounds (cases 1-10, 11-25, 26-50, 51-76, 76-100). Three novices segmented only 50 cases (three rounds). After each round, a one-on-one feedback session between the expert and the reader was held, with feedback on systematic errors and potential improvements for the next round. Dice similarity coefficient (DSC) > 0.8 was considered accurate. RESULTS: Using DSC > 0.8 as the threshold, the novices had a total of 194 accurate segmentations out of 250 (77.6%). The residents had a total of 397/400 (99.2%) accurate segmentations. In round 1, the novices had 19/40 (47.5%) accurate segmentations, in round 2 41/60 (68.3%), and in round 3 84/100 (84.0%) indicating learning progress. CONCLUSIONS: Radiology residents, regardless of prior experience, showed high segmentation accuracy. Novices showed larger interindividual variation and lower segmentation accuracy than radiology residents. To prepare datasets for artificial intelligence (AI) development, employing radiology residents seems safe and provides a good balance between cost-effectiveness and segmentation accuracy. Employing novices should only be considered on an individual basis. CLINICAL RELEVANCE STATEMENT: Employing radiology residents for prostate MRI segmentation seems safe and can potentially reduce the workload of expert radiologists. Employing novices should only be considered on an individual basis. KEY POINTS: • Using less experienced readers for prostate MRI segmentation is cost-effective but may reduce quality. • Radiology residents provided high accuracy segmentations while novices showed large inter-reader variability. • To prepare datasets for AI development, employing radiology residents seems safe and might provide a good balance between cost-effectiveness and segmentation accuracy while novices should only be employed on an individual basis.

Automated quantification of PET/CT skeletal tumor burden in prostate cancer using artificial intelligence: The PET index.

PURPOSE: Consistent assessment of bone metastases is crucial for patient management and clinical trials in prostate cancer (PCa). We aimed to develop a fully automated convolutional neural network (CNN)-based model for calculating PET/CT skeletal tumor burden in patients with PCa. METHODS: A total of 168 patients from three centers were divided into training, validation, and test groups. Manual annotations of skeletal lesions in [18F]fluoride PET/CT scans were used to train a CNN. The AI model was evaluated in 26 patients and compared to segmentations by physicians and to a SUV 15 threshold. PET index representing the percentage of skeletal volume taken up by lesions was estimated. RESULTS: There was no case in which all readers agreed on prevalence of lesions that the AI model failed to detect. PET index by the AI model correlated moderately strong to physician PET index (mean r = 0.69). Threshold PET index correlated fairly with physician PET index (mean r = 0.49). The sensitivity for lesion detection was 65-76% for AI, 68-91% for physicians, and 44-51% for threshold depending on which physician was considered reference. CONCLUSION: It was possible to develop an AI-based model for automated assessment of PET/CT skeletal tumor burden. The model's performance was superior to using a threshold and provides fully automated calculation of whole-body skeletal tumor burden. It could be further developed to apply to different radiotracers. Objective scan evaluation is a first step toward developing a PET/CT imaging biomarker for PCa skeletal metastases.

Aortic wall segmentation in 18F-sodium fluoride PET/CT scans: Head-to-head comparison of artificial intelligence-based versus manual segmentation.

BACKGROUND: We aimed to establish and test an automated AI-based method for rapid segmentation of the aortic wall in positron emission tomography/computed tomography (PET/CT) scans. METHODS: For segmentation of the wall in three sections: the arch, thoracic, and abdominal aorta, we developed a tool based on a convolutional neural network (CNN), available on the Research Consortium for Medical Image Analysis (RECOMIA) platform, capable of segmenting 100 different labels in CT images. It was tested on 18F-sodium fluoride PET/CT scans of 49 subjects (29 healthy controls and 20 angina pectoris patients) and compared to data obtained by manual segmentation. The following derived parameters were compared using Bland-Altman Limits of Agreement: segmented volume, and maximal, mean, and total standardized uptake values (SUVmax, SUVmean, SUVtotal). The repeatability of the manual method was examined in 25 randomly selected scans. RESULTS: CNN-derived values for volume, SUVmax, and SUVtotal were all slightly, i.e., 13-17%, lower than the corresponding manually obtained ones, whereas SUVmean values for the three aortic sections were virtually identical for the two methods. Manual segmentation lasted typically 1-2 hours per scan compared to about one minute with the CNN-based approach. The maximal deviation at repeat manual segmentation was 6%. CONCLUSIONS: The automated CNN-based approach was much faster and provided parameters that were about 15% lower than the manually obtained values, except for SUVmean values, which were comparable. AI-based segmentation of the aorta already now appears as a trustworthy and fast alternative to slow and cumbersome manual segmentation.

"Global" cardiac atherosclerotic burden assessed by artificial intelligence-based versus manual segmentation in 18F-sodium fluoride PET/CT scans: Head-to-head comparison.

BACKGROUND: Artificial intelligence (AI) is known to provide effective means to accelerate and facilitate clinical and research processes. So in this study it was aimed to compare a AI-based method for cardiac segmentation in positron emission tomography/computed tomography (PET/CT) scans with manual segmentation to assess global cardiac atherosclerosis burden. METHODS: A trained convolutional neural network (CNN) was used for cardiac segmentation in 18F-sodium fluoride PET/CT scans of 29 healthy volunteers and 20 angina pectoris patients and compared with manual segmentation. Parameters for segmented volume (Vol) and mean, maximal, and total standardized uptake values (SUVmean, SUVmax, SUVtotal) were analyzed by Bland-Altman Limits of Agreement. Repeatability with AI-based assessment of the same scans is 100%. Repeatability (same conditions, same operator) and reproducibility (same conditions, two different operators) of manual segmentation was examined by re-segmentation in 25 randomly selected scans. RESULTS: Mean (± SD) values with manual vs. CNN-based segmentation were Vol 617.65 ± 154.99 mL vs 625.26 ± 153.55 mL (P = .21), SUVmean 0.69 ± 0.15 vs 0.69 ± 0.15 (P = .26), SUVmax 2.68 ± 0.86 vs 2.77 ± 1.05 (P = .34), and SUVtotal 425.51 ± 138.93 vs 427.91 ± 132.68 (P = .62). Limits of agreement were - 89.42 to 74.2, - 0.02 to 0.02, - 1.52 to 1.32, and - 68.02 to 63.21, respectively. Manual segmentation lasted typically 30 minutes vs about one minute with the CNN-based approach. The maximal deviation at manual re-segmentation was for the four parameters 0% to 0.5% with the same and 0% to 1% with different operators. CONCLUSION: The CNN-based method was faster and provided values for Vol, SUVmean, SUVmax, and SUVtotal comparable to the manually obtained ones. This AI-based segmentation approach appears to offer a more reproducible and much faster substitute for slow and cumbersome manual segmentation of the heart.

Brain tumor segmentation using synthetic MR images - A comparison of GANs and diffusion models.

Large annotated datasets are required for training deep learning models, but in medical imaging data sharing is often complicated due to ethics, anonymization and data protection legislation. Generative AI models, such as generative adversarial networks (GANs) and diffusion models, can today produce very realistic synthetic images, and can potentially facilitate data sharing. However, in order to share synthetic medical images it must first be demonstrated that they can be used for training different networks with acceptable performance. Here, we therefore comprehensively evaluate four GANs (progressive GAN, StyleGAN 1-3) and a diffusion model for the task of brain tumor segmentation (using two segmentation networks, U-Net and a Swin transformer). Our results show that segmentation networks trained on synthetic images reach Dice scores that are 80%-90% of Dice scores when training with real images, but that memorization of the training images can be a problem for diffusion models if the original dataset is too small. Our conclusion is that sharing synthetic medical images is a viable option to sharing real images, but that further work is required. The trained generative models and the generated synthetic images are shared on AIDA data hub.

Improving sensitivity through data augmentation with synthetic lymph node metastases for AI-based analysis of PSMA PET-CT images.

BACKGROUND: We developed a fully automated artificial intelligence (AI)AI-based-based method for detecting suspected lymph node metastases in prostate-specific membrane antigen (PSMA)(PSMA) positron emission tomography-computed tomography (PET-CT)(PET-CT) images of prostate cancer patients by using data augmentation that adds synthetic lymph node metastases to the images to expand the training set. METHODS: Synthetic data were derived from original training images to which new synthetic lymph node metastases were added. Thus, the original training set from a previous study (n = 420) was expanded by one synthetic image for every original image (n = 840), which was used to train an AI model. The performance of the AI model was compared to that of nuclear medicine physicians and a previously developed AI model. The human readers were alternately used as a reference and compared to either another reading or AI model. RESULTS: The new AI model had an average sensitivity of 84% for detecting lymph node metastases compared with 78% for human readings. Our previously developed AI method without synthetic data had an average sensitivity of 79%. The number of false positive lesions were slightly higher for the new AI model (average 3.3 instances per patient) compared to human readings and the previous AI model (average 2.8 instances per patient), while the number of false negative lesions was lower. CONCLUSIONS: Creating synthetic lymph node metastases, as a form of data augmentation, on [18F]PSMA-1007F]PSMA-1007 PETPET-CT-CT images improved the sensitivity of an AI model for detecting suspected lymph node metastases. However, the number of false positive lesions increased somewhat.

NoiseNet, a fully automatic noise assessment tool that can identify non-diagnostic CCTA examinations.

Image noise and vascular attenuation are important factors affecting image quality and diagnostic accuracy of coronary computed tomography angiography (CCTA). The aim of this study was to develop an algorithm that automatically performs noise and attenuation measurements in CCTA and to evaluate the ability of the algorithm to identify non-diagnostic examinations. The algorithm, "NoiseNet", was trained and tested on 244 CCTA studies from the Swedish CArdioPulmonary BioImage Study. The model is a 3D U-Net that automatically segments the aortic root and measures attenuation (Hounsfield Units, HU), noise (standard deviation of HU, HUsd) and signal-to-noise ratio (SNR, HU/HUsd) in the aortic lumen, close to the left coronary ostium. NoiseNet was then applied to 529 CCTA studies previously categorized into three subgroups: fully diagnostic, diagnostic with excluded parts and non-diagnostic. There was excellent correlation between NoiseNet and manual measurements of noise (r = 0.948; p < 0.001) and SNR (r = 0.948; <0.001). There was a significant difference in noise levels between the image quality subgroups: fully diagnostic 33.1 (29.8-37.9); diagnostic with excluded parts 36.1 (31.5-40.3) and non-diagnostic 42.1 (35.2-47.7; p < 0.001). Corresponding values for SNR were 16.1 (14.0-18.0); 14.0 (12.4-16.2) and 11.1 (9.6-14.0; p < 0.001). ROC analysis for prediction of a non-diagnostic study showed an AUC for noise of 0.73 (CI 0.64-0.83) and for SNR of 0.80 (CI 0.71-0.89). In conclusion, NoiseNet can perform noise and SNR measurements with high accuracy. Noise and SNR impact image quality and automatic measurements may be used to identify CCTA studies with low image quality.

Pre-diabetes is associated with attenuation rather than volume of epicardial adipose tissue on computed tomography.

The volume of epicardial adipose tissue (EATV) is increased in type-2 diabetes (T2D), while its attenuation (EATA) appears to be decreased. Similar patterns have been suggested in pre-diabetes, but data is scarce. In both pre-diabetes and T2D, any independent role of EATV and EATA in disease development remains to be proven, a task complicated by their substantial co-variation with other anthropometrics, e.g. BMI, waist circumference, and abdominal visceral adipose tissue (VAT). EATV and EATA was quantified in computed tomography (CT) images in a population study (n = 1948) using an automatic technique. Data was available on BMI, waist circumference, abdominal visceral adipose tissue (VAT) area, insulin resistance (IR) and glucose tolerance, the latter ranging from normal (NGT), over pre-diabetes (impaired fasting glucose [IFG, n = 414] impaired glucose tolerance [IGT, n = 321] and their combination [CGI, n = 128]), to T2D. EATV was increased in pre-diabetes, T2D and IR in univariable analyses and when adjusting for BMI, however not when adjusting for waist or VAT. EATA was reduced in pre-diabetes, T2D and IR in univariable analyses and when adjusting for BMI and waist, however not when adjusting for VAT. Adjustment for other co-variates had little influence on the results. In conclusion, EATV is increased and EATA reduced in pre-diabetes, T2D and IR, however, significant co-variation with other anthropometrics, especially VAT, obscures their function in disease development. The current results do not exclude a pathophysiological role of epicardial fat, but future studies need to adjust for anthropometrics, or focus on the microenvironment within the pericardial sac.

Common carotid segmentation in 18 F-sodium fluoride PET/CT scans: Head-to-head comparison of artificial intelligence-based and manual method.

BACKGROUND: Carotid atherosclerosis is a major cause of stroke, traditionally diagnosed late. Positron emission tomography/computed tomography (PET/CT) with 18 F-sodium fluoride (NaF) detects arterial wall micro-calcification long before macro-calcification becomes detectable by ultrasound, CT or magnetic resonance imaging. However, manual PET/CT processing is time-consuming and requires experience. We compared a convolutional neural network (CNN) approach with manual segmentation of the common carotids. METHODS: Segmentation in NaF-PET/CT scans of 29 healthy volunteers and 20 angina pectoris patients were compared for segmented volume (Vol) and mean, maximal, and total standardized uptake values (SUVmean, SUVmax, and SUVtotal). SUVmean was the average of SUVmeans within the VOI, SUVmax the highest SUV in all voxels in the VOI, and SUVtotal the SUVmean multiplied by the Vol of the VOI. Intra and Interobserver variability with manual segmentation was examined in 25 randomly selected scans. RESULTS: Bias for Vol, SUVmean, SUVmax, and SUVtotal were 1.33 ± 2.06, -0.01 ± 0.05, 0.09 ± 0.48, and 1.18 ± 1.99 in the left and 1.89 ± 1.5, -0.07 ± 0.12, 0.05 ± 0.47, and 1.61 ± 1.47, respectively, in the right common carotid artery. Manual segmentation lasted typically 20 min versus 1 min with the CNN-based approach. Mean Vol deviation at repeat manual segmentation was 14% and 27% in left and right common carotids. CONCLUSIONS: CNN-based segmentation was much faster and provided SUVmean values virtually identical to manually obtained ones, suggesting CNN-based analysis as a promising substitute of slow and cumbersome manual processing.

PET/CT imaging of spinal inflammation and microcalcification in patients with low back pain: A pilot study on the quantification by artificial intelligence-based segmentation.

BACKGROUND: Current imaging modalities are often incapable of identifying nociceptive sources of low back pain (LBP). We aimed to characterize these by means of positron emission tomography/computed tomography (PET/CT) of the lumbar spine region applying tracers 18 F-fluorodeoxyglucose (FDG) and 18 F-sodium fluoride (NaF) targeting inflammation and active microcalcification, respectively. METHODS: Using artificial intelligence (AI)-based quantification, we compared PET findings in two sex- and age-matched groups, a case group of seven males and five females, mean age 45 ± 14 years, with ongoing LBP and a similar control group of 12 pain-free individuals. PET/CT scans were segmented into three distinct volumes of interest (VOIs): lumbar vertebral bodies, facet joints and intervertebral discs. Maximum, mean and total standardized uptake values (SUVmax, SUVmean and SUVtotal) for FDG and NaF uptake in the 3 VOIs were measured and compared between groups. Holm-Bonferroni correction was applied to adjust for multiple testing. RESULTS: FDG uptake was slightly higher in most locations of the LBP group including higher SUVmean in the intervertebral discs (0.96 ± 0.34 vs. 0.69 ± 0.15). All NaF uptake values were higher in cases, including higher SUVmax in the intervertebral discs (11.63 ± 3.29 vs. 9.45 ± 1.32) and facet joints (14.98 ± 6.55 vs. 10.60 ± 2.97). CONCLUSION: Observed intergroup differences suggest acute inflammation and microcalcification as possible nociceptive causes of LBP. AI-based quantification of relevant lumbar VOIs in PET/CT scans of LBP patients and controls appears to be feasible. These promising, early findings warrant further investigation and confirmation.

Artificial intelligence based automatic quantification of epicardial adipose tissue suitable for large scale population studies.

To develop a fully automatic model capable of reliably quantifying epicardial adipose tissue (EAT) volumes and attenuation in large scale population studies to investigate their relation to markers of cardiometabolic risk. Non-contrast cardiac CT images from the SCAPIS study were used to train and test a convolutional neural network based model to quantify EAT by: segmenting the pericardium, suppressing noise-induced artifacts in the heart chambers, and, if image sets were incomplete, imputing missing EAT volumes. The model achieved a mean Dice coefficient of 0.90 when tested against expert manual segmentations on 25 image sets. Tested on 1400 image sets, the model successfully segmented 99.4% of the cases. Automatic imputation of missing EAT volumes had an error of less than 3.1% with up to 20% of the slices in image sets missing. The most important predictors of EAT volumes were weight and waist, while EAT attenuation was predicted mainly by EAT volume. A model with excellent performance, capable of fully automatic handling of the most common challenges in large scale EAT quantification has been developed. In studies of the importance of EAT in disease development, the strong co-variation with anthropometric measures needs to be carefully considered.

Artificial intelligence-based detection of lymph node metastases by PET/CT predicts prostate cancer-specific survival.

INTRODUCTION: Lymph node metastases are a key prognostic factor in prostate cancer (PCa), but detecting lymph node lesions from PET/CT images is a subjective process resulting in inter-reader variability. Artificial intelligence (AI)-based methods can provide an objective image analysis. We aimed at developing and validating an AI-based tool for detection of lymph node lesions. METHODS: A group of 399 patients with biopsy-proven PCa who had undergone 18 F-choline PET/CT for staging prior to treatment were used to train (n = 319) and test (n = 80) the AI-based tool. The tool consisted of convolutional neural networks using complete PET/CT scans as inputs. In the test set, the AI-based lymph node detections were compared to those of two independent readers. The association with PCa-specific survival was investigated. RESULTS: The AI-based tool detected more lymph node lesions than Reader B (98 vs. 87/117; p = .045) using Reader A as reference. AI-based tool and Reader A showed similar performance (90 vs. 87/111; p = .63) using Reader B as reference. The number of lymph node lesions detected by the AI-based tool, PSA, and curative treatment was significantly associated with PCa-specific survival. CONCLUSION: This study shows the feasibility of using an AI-based tool for automated and objective interpretation of PET/CT images that can provide assessments of lymph node lesions comparable with that of experienced readers and prognostic information in PCa patients.

Pneumococcal carriage among children aged 4 - 12 years in Angola 4 years after the introduction of a pneumococcal conjugate vaccine.

Children in Angola are affected by a high burden of disease caused by pneumococcal infections. The 13-valent pneumococcal conjugate vaccine (PCV13) was introduced in the childhood immunization programme in 2013 but the serotype distribution of Streptococcus pneumoniae and antimicrobial susceptibility patterns are unknown. We did a cross-sectional nasopharyngeal carriage study in Luanda and Saurimo, Angola (PCV13 3rd dose coverage 67% and 84%, respectively) during November to December 2017 comprising 940 children aged 4-12 years. The main objective was to assess vaccine serotype coverage and antimicrobial susceptibility rates for S. pneumoniae. Our secondary aim was to characterize colonizinig strains of Haemophilus influenzae and Moraxella catarrhalis. Pneumococcal colonization was found in 35% (95% CI 32-39%) of children (n = 332), with 41% of serotypes covered by PCV13. The most common serotypes were 3 (8%), 18C (6%), 23F (6%), 11A (6%), 34 (6%), 19F (5%) and 16 (5%). Carriage of H. influenzae and M. catarrhalis was detected in 13% (95% CI 11-15%) and 15% (95% CI 13-17%) of children, respectively. Non-susceptibility to penicillin was common among pneumococci (40%), particularly among PCV13-included serotypes (50% vs. 33%; p = 0.003), although the median minimal inhibitory concentration was low (0.19 µg/mL, IQR 0.13-0.25 µg/mL). Most pneumococci and H. influenzae were susceptible to amoxicillin (99% and 88%, respectively). Furthermore, resistance to trimethoprim-sulfamethoxazole was>70% among all three species. Multidrug-resistant pneumococci (non-susceptible to ≥ 3 antibiotics; 7% [n = 24]) were further studied with whole genome sequencing to investigate clonality as an underlying cause for this phenotype. No clearly dominating clone(s) were, however, detected. The results indicate that continued use of PCV13 may have positive direct and herd effects on pneumococcal infections in Angola as carriage of vaccine serotypes was common in the non-vaccinated age group. Finally, amoxicillin is assessed to be a feasible empirical treatment of respiratory tract infections in Angola.