Automated Deep Learning Segmentation of Cardiac Inflammatory FDG PET.

BACKGROUND: Fluorodeoxyglucose positron emission tomography (FDG PET) with suppression of myocardial glucose utilization plays a pivotal role in diagnosing cardiac sarcoidosis. Reorientation of images to match perfusion datasets and myocardial segmentation enables consistent image scaling and quantification. However, such manual tasks are cumbersome. We developed a 3D U-Net deep-learning (DL) algorithm for automated myocardial segmentation in cardiac sarcoidosis FDG PET. METHODS: The DL model was trained on FDG PET scans from 316 patients with left ventricular contours derived from paired perfusion datasets. Qualitative analysis of clinical readability was performed to compare DL segmentation with the current automated method on a 50-patient test subset. Additionally, left ventricle displacement and angulation, as well as SUVmax sampling were compared to inter-user reproducibility results. A hybrid workflow was also investigated to accelerate study processing time. RESULTS: DL segmentation enhanced readability scores in over 90% of cases compared to the standard segmentation currently used in the software. DL segmentation performed similarly to a trained technologist, surpassing standard segmentation for left ventricle displacement and angulation, as well as correlation of SUVmax. Using the DL segmentation as initial placement for manual segmentation significantly decreased processing time. CONCLUSION: A novel DL-based automated segmentation tool markedly improves processing of cardiac sarcoidosis FDG PET. This tool yields optimized splash display of sarcoidosis FDG PET datasets with no user input and offers significant processing time improvement for manual segmentation of such datasets.

Automated Deep Learning Segmentation of Cardiac Inflammatory FDG PET.

Background: Fluorodeoxyglucose positron emission tomography (FDG PET) with glycolytic metabolism suppression plays a pivotal role in diagnosing cardiac sarcoidosis. Reorientation of images to match perfusion datasets is critical and myocardial segmentation enables consistent image scaling and quantification. However, both are challenging and labor intensive. We developed a 3D U-Net deep learning (DL) algorithm for automated myocardial segmentation in cardiac sarcoidosis FDG PET. Methods: The DL model was trained on 316 patients' FDG PET scans, and left ventricular contours derived from perfusion datasets. Qualitative analysis of clinical readability was performed to compare DL segmentation with the current automated method on a 50-patient test subset. Additionally, left ventricle displacement and angulation, as well as SUVmax sampling were compared to inter-user reproducibility results. Results: DL segmentation enhanced readability scores in over 90% of cases compared to the standard segmentation currently used in the software. DL segmentation performed similarly to a trained technologist, surpassing standard segmentation for left ventricle displacement and angulation, as well as correlation of SUVmax. Conclusion: The DL-based automated segmentation tool presents a marked improvement in the processing of cardiac sarcoidosis FDG PET, promising enhanced clinical workflow. This tool holds significant potential for accelerating clinical practice and improving consistency and quality. Further research with varied datasets is warranted to broaden its applicability.

Integrated myocardial flow reserve (iMFR) assessment: optimized PET blood flow quantification for diagnosis of coronary artery disease.

PURPOSE: Distinguishing obstructive epicardial coronary artery disease (CAD) from microvascular dysfunction and diffuse atherosclerosis would be of immense benefit clinically. However, quantitative measures of absolute myocardial blood flow (MBF) integrate the effects of focal epicardial stenosis, diffuse atherosclerosis, and microvascular dysfunction. In this study, MFR and relative perfusion quantification were combined to create integrated MFR (iMFR) which was evaluated using data from a large clinical registry and an international multi-center trial and validated against invasive coronary angiography (ICA). METHODS: This study included 1,044 clinical patients referred for 82Rb rest/stress positron emission tomography myocardial perfusion imaging and ICA, along with 231 patients from the Flurpiridaz 301 trial (clinicaltrials.gov NCT01347710). MFR and relative perfusion quantification were combined to create an iMFR map. The incremental value of iMFR was evaluated for diagnosis of obstructive stenosis, adjusted for patient demographics and pre-test probability of CAD. Models for high-risk anatomy (left main or three-vessel disease) were also constructed. RESULTS: iMFR parameters of focally impaired perfusion resulted in best fitting diagnostic models. Receiver-operating characteristic analysis showed a slight improvement compared to standard quantitative perfusion approaches (AUC 0.824 vs. 0.809). Focally impaired perfusion was also associated with high-risk CAD anatomy (OR 1.40 for extent, and OR 2.40 for decreasing mean MFR). Diffusely impaired perfusion was associated with lower likelihood of obstructive CAD, and, in the absence of transient ischemic dilation (TID), with lower likelihood of high-risk CAD anatomy. CONCLUSIONS: Focally impaired perfusion extent derived from iMFR assessment is a powerful incremental predictor of obstructive CAD while diffusely impaired perfusion extent can help rule out obstructive and high-risk CAD in the absence of TID.

"Virtual" attenuation correction: improving stress myocardial perfusion SPECT imaging using deep learning.

PURPOSE: Myocardial perfusion imaging (MPI) using single-photon emission computed tomography (SPECT) is widely used for coronary artery disease (CAD) evaluation. Although attenuation correction is recommended to diminish image artifacts and improve diagnostic accuracy, approximately 3/4ths of clinical MPI worldwide remains non-attenuation-corrected (NAC). In this work, we propose a novel deep learning (DL) algorithm to provide "virtual" DL attenuation-corrected (DLAC) perfusion polar maps solely from NAC data without concurrent computed tomography (CT) imaging or additional scans. METHODS: SPECT MPI studies (N = 11,532) with paired NAC and CTAC images were retrospectively identified. A convolutional neural network-based DL algorithm was developed and trained on half of the population to predict DLAC polar maps from NAC polar maps. Total perfusion deficit (TPD) was evaluated for all polar maps. TPDs from NAC and DLAC polar maps were compared to CTAC TPDs in linear regression analysis. Moreover, receiver-operating characteristic analysis was performed on NAC, CTAC, and DLAC TPDs to predict obstructive CAD as diagnosed from invasive coronary angiography. RESULTS: DLAC TPDs exhibited significantly improved linear correlation (p < 0.001) with CTAC (R2 = 0.85) compared to NAC vs. CTAC (R2 = 0.68). The diagnostic performance of TPD was also improved with DLAC compared to NAC with an area under the curve (AUC) of 0.827 vs. 0.780 (p = 0.012) with no statistically significant difference between AUC for CTAC and DLAC. At 88% sensitivity, specificity was improved by 18.9% for DLAC and 25.6% for CTAC. CONCLUSIONS: The proposed DL algorithm provided attenuation correction comparable to CTAC without the need for additional scans. Compared to conventional NAC perfusion imaging, DLAC significantly improved diagnostic accuracy.

Effect of iterations and time of flight on normal distributions of 82Rb PET relative perfusion and myocardial blood flow.

BACKGROUND: As clinical use of myocardial blood flow (MBF) increases, dynamic series are becoming part of the typical workflow. The methods and parameters used to reconstruct these series require investigation to ensure accurate quantification. METHODS: Fifty-nine rest/stress dynamic 82Rb PET studies, acquired on a Biograph mCT, from a combination of normal volunteers and low-likelihood patients were reconstructed with and without time of flight (TOF) for varying iterations and processed to obtain relative perfusion and MBF polar maps. Regional values from mean polar maps were fit to a linear mixed-effect model to quantify convergence and select the optimal number of iterations. RESULTS: TOF reconstructions converged faster and yielded more uniform relative perfusion polar maps. However, the stress MBF distribution for TOF reconstructions was more heterogeneous, with a higher-intensity septal wall. This phenomenon requires further investigation, with right ventricle blood pool spillover possibly having an effect. Optimal reconstructions were defined as 5-iteration non-TOF (24-subset) reconstructions and 3-iteration TOF (21-subset) reconstructions. CONCLUSION: Optimal cardiac reconstructions were identified for non-TOF and TOF reconstructions of dynamic series. TOF reconstruction presents as the more accurate method, given the more uniform relative perfusion distribution.

Impact of residual subtraction on myocardial blood flow and reserve estimates from rapid dynamic PET protocols.

BACKGROUND: 13N-ammonia and 18F-flurpiridaz require longer delays between rest and stress studies to allow for decay, lowering clinical throughput. In this study, we investigated the impact of residual subtraction on MBF and MFR estimates, as well as its effects on diagnostic accuracy. METHODS: We retrospectively analyzed 63 patients who underwent a dynamic ammonia rest/stress study and 231 patients from the flurpiridaz 301 trial. Residual subtraction was performed by subtracting the mean pre-injection activity in each sampled region from that region's time activity curve. Corrected and uncorrected MBF and MFR were analyzed. Diagnostic accuracy was compared to quantitative coronary angiograms (QCA) for the flurpiridaz population. RESULTS: With delays between injections above 3 half-lives, and a doubled stress dose, residual activity did not meaningfully increase ammonia MBF (< 5%). For shorter injection delays, stress MBF was overestimated by 13.6% ± 5.0% (P < .001). Residual activity had a large effect on flurpiridaz stress MBF, overestimating it by 37.9% ± 23.2% (P < .001). Comparison to QCA showed a significant improvement in AUC with residual subtraction (from 0.748 to 0.831, P = .001). MFR yielded similar results. CONCLUSIONS: Accounting for residual activity has a marked impact on stress MBF and MFR and improves diagnostic accuracy relative to QCA.

Myocardial flow reserve estimation with contemporary CZT-SPECT and 99mTc-tracers lacks precision for routine clinical application.

BACKGROUND: PET myocardial flow reserve (MFR) has established diagnostic and prognostic value. Technological advances have now enabled SPECT MFR quantification. We investigated whether SPECT MFR precision is sufficient for clinical categorization of patients. METHODS: Validation studies vs invasive flow measurements and PET MFR were reviewed to determine global SPECT MFR thresholds. Studies vs PET and a SPECT MFR repeatability study were used to establish imprecision in SPECT MFR measurements as the standard deviation of the difference between SPECT and PET MFR, or test-retest SPECT MFR. Simulations were used to evaluate the impact of SPECT MFR imprecision on confidence of clinically relevant categorization. RESULTS: Based on validation studies, the typical PET MFR categories were used for SPECT MFR classification (< 1.5, 1.5-2.0, > 2.0). Imprecision vs PET MFR ranged from 0.556 to 0.829, and test-retest imprecision was 0.781-0.878. Simulations showed correct classification of up to only 34% of patients when 1.5 ≤ true MFR ≤ 2.0. Categorization with high confidence (> 80%) was only achieved for extreme MFR values (< 1.0 or > 2.5), with correct classification in only 15% of patients in a typical lab with MFR of 1.8 ± 0.5. CONCLUSIONS: Current SPECT-derived estimates of MFR lack precision and require further optimization for clinical risk stratification.

Effects of two patient-specific dosing protocols on measurement of myocardial blood flow with 3D 82Rb cardiac PET.

PURPOSE: Clinical measurement of myocardial blood flow (MBF) has emerged as an important component of routine PET-CT assessment of myocardial perfusion in patients with known or suspected coronary artery disease. Although multiple society guidelines recommend patient-specific dosing, there is a lack of studies evaluating the efficacy of patient-specific dosing for quantitative MBF accuracy. METHODS: Two patient-specific dosing protocols (weight- and BMI-adjusted) were retrospectively evaluated in 435 consecutive clinical patients referred for PET myocardial perfusion assessment. MBF was estimated at rest and after regadenoson-induced hyperemia. The effect of dosing protocol on dose reduction, PET scanner saturation, relative perfusion, and image quality was compared. The effect of PET saturation on the accuracy of MBF and myocardial flow reserve (MFR) in remote myocardium was assessed with multivariable linear regression. RESULTS: BMI-adjusted dosing was associated with lower administered 82Rb activities (1036.0 ± 274 vs. 1147 ± 274 MBq, p = 0.003) and lower PET scanner saturation incidence (28 vs. 38%, p = 0.006) and severity (median saturation severity index 0.219 ± 0.33 vs. 0.397 ± 0.59%, p = 0.018) compared to weight-adjusted dosing. PET saturation that occurred with either dosing protocol was moderate and resulted in modest remote MBF and MFR biases ranging from 2 to 9% after adjusting for patient age, sex, BMI, rate-pressure product, and LV ejection fraction. No adverse effects of BMI dose adjustment were observed in relative perfusion assessment or image quality. CONCLUSIONS: Patient-specific dosing according to BMI is an effective method for guideline-directed dose reduction while maintaining image quality and accuracy for routine MBF and MFR quantification.