Amyloid PET Quantification Via End-to-End Training of a Deep Learning

Multi-instance learning attention model for amyloid quantification of brain sub regions in longitudinal cognitive decline

Amyloid PET scans help in identifying the beta-amyloid deposition in different brain regions. The purpose of this study is to develop a deep learning model that can automate the task of finding amyloid deposition in different regions of the brain only by using PET scan and without the corresponding MRI scan. 2647 18F-Florbetapir PET scans are collected from Alzheimer's Disease Neuroimaging Initiative (ADNI) from multiple centres taken over a period. A deep learning model based on multi-instance learning and attention is proposed which is trained and validated using 80% of the scans and the remaining 20% of the scans are used for testing the model. The performance of the model is validated using Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). The proposed model is further tested upon an external dataset consisting of 1413 18F-Florbetapir PET scans from the Anti-Amyloid Treatment in Asymptomatic Alzheimer's (A4) study. The proposed model achieves MAE of 0.0243 and RMSE of 0.0320 for summary Standardized Uptake Value Ratio (SUVR) based on composite reference region for ADNI test set. When tested on the A4-study dataset, the proposed model achieves MAE of 0.038 and RMSE of 0.0495 for summary SUVR based on the composite region. The results show that the proposed model provides less MAE and RMSE when compared with existing models. A graphical user interface is developed based on the proposed model where the predictions are made by selecting the files of 18F-Florbetapir PET scans.

Generative AI unlocks PET insights: brain amyloid dynamics and quantification

Introduction

Studying the spatiotemporal patterns of amyloid accumulation in the brain over time is crucial in understanding Alzheimer's disease (AD). Positron Emission Tomography (PET) imaging plays a pivotal role because it allows for the visualization and quantification of abnormal amyloid beta (Aβ) load in the living brain, providing a powerful tool for tracking disease progression and evaluating the efficacy of anti-amyloid therapies. Generative artificial intelligence (AI) can learn complex data distributions and generate realistic synthetic images. In this study, we demonstrate for the first time the potential of Generative Adversarial Networks (GANs) to build a low-dimensional representation space that effectively describes brain amyloid load and its dynamics.

Methods

Using a cohort of 1,259 subjects with AV45 PET images from the Alzheimer's Disease Neuroimaging Initiative (ADNI), we develop a 3D GAN model to project images into a latent representation space and generate back synthetic images. Then, we build a progression model on the representation space based on non-parametric ordinary differential equations to study brain amyloid evolution.

Results

We found that global SUVR can be accurately predicted with a linear regression model only from the latent representation space (RMSE = 0.08 ± 0.01). We generated synthetic PET trajectories and illustrated predicted Aβ change in four years compared with actual progression.

Discussion

Generative AI can generate rich representations for statistical prediction and progression modeling and simulate evolution in synthetic patients, providing an invaluable tool for understanding AD, assisting in diagnosis, and designing clinical trials. The aim of this study was to illustrate the huge potential that generative AI has in brain amyloid imaging and to encourage its advancement by providing use cases and ideas for future research tracks.

AmyloidPETNet: Classification of Amyloid Positivity in Brain PET Imaging Using End-to-End Deep Learning

Background Visual assessment of amyloid PET scans relies on the availability of radiologist expertise, whereas quantification of amyloid burden typically involves MRI for processing and analysis, which can be computationally expensive. Purpose To develop a deep learning model to classify minimally processed brain PET scans as amyloid positive or negative, evaluate its performance on independent data sets and different tracers, and compare it with human visual reads. Materials and Methods This retrospective study used 8476 PET scans (6722 patients) obtained from late 2004 to early 2023 that were analyzed across five different data sets. A deep learning model, AmyloidPETNet, was trained on 1538 scans from 766 patients, validated on 205 scans from 95 patients, and internally tested on 184 scans from 95 patients in the Alzheimer's Disease Neuroimaging Initiative (ADNI) fluorine 18 (18F) florbetapir (FBP) data set. It was tested on ADNI scans using different tracers and scans from independent data sets. Scan amyloid positivity was based on mean cortical standardized uptake value ratio cutoffs. To compare with model performance, each scan from both the Centiloid Project and a subset of the Anti-Amyloid Treatment in Asymptomatic Alzheimer's Disease (A4) study were visually interpreted with a confidence level (low, intermediate, high) of amyloid positivity/negativity. The area under the receiver operating characteristic curve (AUC) and other performance metrics were calculated, and Cohen κ was used to measure physician-model agreement. Results The model achieved an AUC of 0.97 (95% CI: 0.95, 0.99) on test ADNI 18F-FBP scans, which generalized well to 18F-FBP scans from the Open Access Series of Imaging Studies (AUC, 0.95; 95% CI: 0.93, 0.97) and the A4 study (AUC, 0.98; 95% CI: 0.98, 0.98). Model performance was high when applied to data sets with different tracers (AUC ≥ 0.97). Other performance metrics provided converging evidence. Physician-model agreement ranged from fair (Cohen κ = 0.39; 95% CI: 0.16, 0.60) on a sample of mostly equivocal cases from the A4 study to almost perfect (Cohen κ = 0.93; 95% CI: 0.86, 1.0) on the Centiloid Project. Conclusion The developed model was capable of automatically and accurately classifying brain PET scans as amyloid positive or negative without relying on experienced readers or requiring structural MRI. Clinical trial registration no. NCT00106899 © RSNA, 2024 Supplemental material is available for this article. See also the editorial by Bryan and Forghani in this issue.

ArcheD, a residual neural network for prediction of cerebrospinal fluid amyloid-beta from amyloid PET images

Detection and measurement of amyloid-beta (Aβ) in the brain is a key factor for early identification and diagnosis of Alzheimer's disease (AD). We aimed to develop a deep learning model to predict Aβ cerebrospinal fluid (CSF) concentration directly from amyloid PET images, independent of tracers, brain reference regions or preselected regions of interest. We used 1870 Aβ PET images and CSF measurements to train and validate a convolutional neural network ("ArcheD"). We evaluated the ArcheD performance in relation to episodic memory and the standardized uptake value ratio (SUVR) of cortical Aβ. We also compared the brain region's relevance for the model's CSF prediction within clinical-based and biological-based classifications. ArcheD-predicted Aβ CSF values correlated with measured Aβ CSF values (r = 0.92; q < 0.01), SUVR (rAV45 = -0.64, rFBB = -0.69; q < 0.01) and episodic memory measures (0.33 < r < 0.44; q < 0.01). For both classifications, cerebral white matter significantly contributed to CSF prediction (q < 0.01), specifically in non-symptomatic and early stages of AD. However, in late-stage disease, the brain stem, subcortical areas, cortical lobes, limbic lobe and basal forebrain made more significant contributions (q < 0.01). Considering cortical grey matter separately, the parietal lobe was the strongest predictor of CSF amyloid levels in those with prodromal or early AD, while the temporal lobe played a more crucial role for those with AD. In summary, ArcheD reliably predicted Aβ CSF concentration from Aβ PET scans, offering potential clinical utility for Aβ level determination.

A Review of Machine Learning Approaches for Brain Positron Emission Tomography Data Analysis

Positron emission tomography (PET) imaging has moved forward the development of medical diagnostics and research across various domains, including cardiology, neurology, infection detection, and oncology. The integration of machine learning (ML) algorithms into PET data analysis has further enhanced their capabilities of including disease diagnosis and classification, image segmentation, and quantitative analysis. ML algorithms empower researchers and clinicians to extract valuable insights from complex big PET datasets, which enabling automated pattern recognition, predictive health outcome modeling, and more efficient data analysis. This review explains the basic knowledge of PET imaging, statistical methods for PET image analysis, and challenges of PET data analysis. We also discussed the improvement of analysis capabilities by combining PET data with machine learning algorithms and the application of this combination in various aspects of PET image research. This review also highlights current trends and future directions in PET imaging, emphasizing the driving and critical role of machine learning and big PET image data analytics in improving diagnostic accuracy and personalized medical approaches. Integration between PET imaging will shape the future of medical diagnosis and research.

Deep Learning-Driven Estimation of Centiloid Scales from Amyloid PET Images with 11C-PiB and 18F-Labeled Tracers in Alzheimer's Disease

BACKGROUND

Standard methods for deriving Centiloid scales from amyloid PET images are time-consuming and require considerable expert knowledge. We aimed to develop a deep learning method of automating Centiloid scale calculations from amyloid PET images with 11C-Pittsburgh Compound-B (PiB) tracer and assess its applicability to 18F-labeled tracers without retraining.

METHODS

We trained models on 231 11C-PiB amyloid PET images using a 50-layer 3D ResNet architecture. The models predicted the Centiloid scale, and accuracy was assessed using mean absolute error (MAE), linear regression analysis, and Bland-Altman plots.

RESULTS

The MAEs for Alzheimer's disease (AD) and young controls (YC) were 8.54 and 2.61, respectively, using 11C-PiB, and 8.66 and 3.56, respectively, using 18F-NAV4694. The MAEs for AD and YC were higher with 18F-florbetaben (39.8 and 7.13, respectively) and 18F-florbetapir (40.5 and 12.4, respectively), and the error rate was moderate for 18F-flutemetamol (21.3 and 4.03, respectively). Linear regression yielded a slope of 1.00, intercept of 1.26, and R2 of 0.956, with a mean bias of -1.31 in the Centiloid scale prediction.

CONCLUSIONS

We propose a deep learning means of directly predicting the Centiloid scale from amyloid PET images in a native space. Transferring the model trained on 11C-PiB directly to 18F-NAV4694 without retraining was feasible.

ArcheD, a residual neural network for prediction of cerebrospinal fluid amyloid-beta from amyloid PET images

Detection and measurement of amyloid-beta (Aβ) aggregation in the brain is a key factor for early identification and diagnosis of Alzheimer's disease (AD). We aimed to develop a deep learning model to predict Aβ cerebrospinal fluid (CSF) concentration directly from amyloid PET images, independent of tracers, brain reference regions or preselected regions of interest. We used 1870 Aβ PET images and CSF measurements to train and validate a convolutional neural network ("ArcheD"). We evaluated the ArcheD performance in relation to episodic memory and the standardized uptake value ratio (SUVR) of cortical Aβ. We also compared the brain region's relevance for the model's CSF prediction within clinical-based and biological-based classifications. ArcheD-predicted Aβ CSF values correlated strongly with measured Aβ CSF values ( r =0.81; p <0.001) and showed correlations with SUVR and episodic memory measures in all participants except in those with AD. For both clinical and biological classifications, cerebral white matter significantly contributed to CSF prediction ( q <0.01), specifically in non-symptomatic and early stages of AD. However, in late-stage disease, brain stem, subcortical areas, cortical lobes, limbic lobe, and basal forebrain made more significant contributions (q<0.01). Considering cortical gray matter separately, the parietal lobe was the strongest predictor of CSF amyloid levels in those with prodromal or early AD, while the temporal lobe played a more crucial role for those with AD. In summary, ArcheD reliably predicted Aβ CSF concentration from Aβ PET scans, offering potential clinical utility for Aβ level determination and early AD detection.

Estimation of brain amyloid accumulation using deep learning in clinical [11C]PiB PET imaging

INTRODUCTION

Estimation of brain amyloid accumulation is valuable for evaluation of patients with cognitive impairment in both research and clinical routine. The development of high throughput and accurate strategies for the determination of amyloid status could be an important tool in patient selection for clinical trials and amyloid directed treatment. Here, we propose the use of deep learning to quantify amyloid accumulation using standardized uptake value ratio (SUVR) and classify amyloid status based on their PET images.

METHODS

A total of 1309 patients with cognitive impairment scanned with [¹¹C]PIB PET/CT or PET/MRI were included. Two convolutional neural networks (CNNs) for reading-based amyloid status and SUVR prediction were trained using 75% of the PET/CT data. The remaining PET/CT (n = 300) and all PET/MRI (n = 100) data was used for evaluation.

RESULTS

The prevalence of amyloid positive patients was 61%. The amyloid status classification model reproduced the expert reader's classification with 99% accuracy. There was a high correlation between reference and predicted SUVR (R² = 0.96). Both reference and predicted SUVR had an accuracy of 97% compared to expert classification when applying a predetermined SUVR threshold of 1.35 for binary classification of amyloid status.

CONCLUSION

The proposed CNN models reproduced both the expert classification and quantitative measure of amyloid accumulation in a large local dataset. This method has the potential to replace or simplify existing clinical routines and can facilitate fast and accurate classification well-suited for a high throughput pipeline.

Automatic Lung Cancer Segmentation in [18F]FDG PET/CT Using a Two-Stage Deep Learning Approach

Purpose

Since accurate lung cancer segmentation is required to determine the functional volume of a tumor in [18F]FDG PET/CT, we propose a two-stage U-Net architecture to enhance the performance of lung cancer segmentation using [18F]FDG PET/CT.

Methods

The whole-body [18F]FDG PET/CT scan data of 887 patients with lung cancer were retrospectively used for network training and evaluation. The ground-truth tumor volume of interest was drawn using the LifeX software. The dataset was randomly partitioned into training, validation, and test sets. Among the 887 PET/CT and VOI datasets, 730 were used to train the proposed models, 81 were used as the validation set, and the remaining 76 were used to evaluate the model. In Stage 1, the global U-net receives 3D PET/CT volume as input and extracts the preliminary tumor area, generating a 3D binary volume as output. In Stage 2, the regional U-net receives eight consecutive PET/CT slices around the slice selected by the Global U-net in Stage 1 and generates a 2D binary image as the output.

Results

The proposed two-stage U-Net architecture outperformed the conventional one-stage 3D U-Net in primary lung cancer segmentation. The two-stage U-Net model successfully predicted the detailed margin of the tumors, which was determined by manually drawing spherical VOIs and applying an adaptive threshold. Quantitative analysis using the Dice similarity coefficient confirmed the advantages of the two-stage U-Net.

Conclusion

The proposed method will be useful for reducing the time and effort required for accurate lung cancer segmentation in [18F]FDG PET/CT.

Voxel-Based Internal Dosimetry for 177Lu-Labeled Radiopharmaceutical Therapy Using Deep Residual Learning

Purpose

In this study, we propose a deep learning (DL)-based voxel-based dosimetry method in which dose maps acquired using the multiple voxel S-value (VSV) approach were used for residual learning.

Methods

Twenty-two SPECT/CT datasets from seven patients who underwent 177Lu-DOTATATE treatment were used in this study. The dose maps generated from Monte Carlo (MC) simulations were used as the reference approach and target images for network training. The multiple VSV approach was used for residual learning and compared with dose maps generated from deep learning. The conventional 3D U-Net network was modified for residual learning. The absorbed doses in the organs were calculated as the mass-weighted average of the volume of interest (VOI).

Results

The DL approach provided a slightly more accurate estimation than the multiple-VSV approach, but the results were not statistically significant. The single-VSV approach yielded a relatively inaccurate estimation. No significant difference was noted between the multiple VSV and DL approach on the dose maps. However, this difference was prominent in the error maps. The multiple VSV and DL approach showed a similar correlation. In contrast, the multiple VSV approach underestimated doses in the low-dose range, but it accounted for the underestimation when the DL approach was applied.

Conclusion

Dose estimation using the deep learning-based approach was approximately equal to that in the MC simulation. Accordingly, the proposed deep learning network is useful for accurate and fast dosimetry after radiation therapy using 177Lu-labeled radiopharmaceuticals.

DeepAD: A deep learning application for predicting amyloid standardized uptake value ratio through PET for Alzheimer's prognosis

Introduction

Amyloid deposition is a vital biomarker in the process of Alzheimer's diagnosis. ¹⁸F-florbetapir PET scans can provide valuable imaging data to determine cortical amyloid quantities. However, the process is labor and doctor intensive, requiring extremely specialized education and resources that may not be accessible to everyone, making the amyloid calculation process inefficient. Deep learning is a rising tool in Alzheimer's research which could be used to determine amyloid deposition.

Materials and methods

Using data from the Alzheimer's Disease Neuroimaging Initiative, we identified 2,980 patients with PET imaging, clinical, and genetic data. We tested various ResNet, EfficientNet, and RegNet convolutional neural networks and later combined the best performing model with Gradient Boosting Decision Tree algorithms to predict standardized uptake value ratio (SUVR) of amyloid in each patient session. We tried several configurations to find the best model tuning for regression-to-SUVR.

Results

We found that the RegNet X064 architecture combined with a grid search-tuned Gradient Boosting Decision Tree with 3 axial input slices and clinical and genetic data achieved the lowest loss. Using the mean-absolute-error metric, the loss converged to an MAE of 0.0441, equating to 96.4% accuracy across the 596-patient test set.

Discussion

We showed that this method is more consistent and accessible in comparison to human readers from previous studies, with lower margins of error and substantially faster calculation times. We implemented our deep learning model on to a web application named DeepAD which allows our diagnostic tool to be accessible. DeepAD could be used in hospitals and clinics with resource limitations for amyloid deposition and shows promise for more imaging tasks as well.

Quantification of amyloid PET for future clinical use: a state-of-the-art review

Amyloid-β (Aβ) pathology is one of the earliest detectable brain changes in Alzheimer's disease (AD) pathogenesis. The overall load and spatial distribution of brain Aβ can be determined in vivo using positron emission tomography (PET), for which three fluorine-18 labelled radiotracers have been approved for clinical use. In clinical practice, trained readers will categorise scans as either Aβ positive or negative, based on visual inspection. Diagnostic decisions are often based on these reads and patient selection for clinical trials is increasingly guided by amyloid status. However, tracer deposition in the grey matter as a function of amyloid load is an inherently continuous process, which is not sufficiently appreciated through binary cut-offs alone. State-of-the-art methods for amyloid PET quantification can generate tracer-independent measures of Aβ burden. Recent research has shown the ability of these quantitative measures to highlight pathological changes at the earliest stages of the AD continuum and generate more sensitive thresholds, as well as improving diagnostic confidence around established binary cut-offs. With the recent FDA approval of aducanumab and more candidate drugs on the horizon, early identification of amyloid burden using quantitative measures is critical for enrolling appropriate subjects to help establish the optimal window for therapeutic intervention and secondary prevention. In addition, quantitative amyloid measurements are used for treatment response monitoring in clinical trials. In clinical settings, large multi-centre studies have shown that amyloid PET results change both diagnosis and patient management and that quantification can accurately predict rates of cognitive decline. Whether these changes in management reflect an improvement in clinical outcomes is yet to be determined and further validation work is required to establish the utility of quantification for supporting treatment endpoint decisions. In this state-of-the-art review, several tools and measures available for amyloid PET quantification are summarised and discussed. Use of these methods is growing both clinically and in the research domain. Concurrently, there is a duty of care to the wider dementia community to increase visibility and understanding of these methods.

Development of Amyloid PET Analysis Pipeline Using Deep Learning-Based Brain MRI Segmentation—A Comparative Validation Study

Amyloid positron emission tomography (PET) scan is clinically essential for the non-invasive assessment of the presence and spatial distribution of amyloid-beta deposition in subjects with cognitive impairment suspected to have been a result of Alzheimer’s disease. Quantitative assessment can enhance the interpretation reliability of PET scan; however, its clinical application has been limited due to the complexity of preprocessing. This study introduces a novel deep-learning-based approach for SUVR quantification that simplifies the preprocessing step and significantly reduces the analysis time. Using two heterogeneous amyloid ligands, our proposed method successfully distinguished standardized uptake value ratio (SUVR) between amyloidosis-positive and negative groups. The proposed method’s intra-class correlation coefficients were 0.97 and 0.99 against PETSurfer and PMOD, respectively. The difference of global SUVRs between the proposed method and PETSurfer or PMOD were 0.04 and −0.02, which are clinically acceptable. The AUC-ROC exceeded 0.95 for three tools in the amyloid positive assessment. Moreover, the proposed method had the fastest processing time and had a low registration failure rate (1%). In conclusion, our proposed method calculates SUVR that is consistent with PETSurfer and PMOD, and has advantages of fast processing time and low registration failure rate. Therefore, PET quantification provided by our proposed method can be used in clinical practice.

Boosting the diagnostic power of amyloid-β PET using a data-driven spatially informed classifier for decision support

BACKGROUND

Amyloid-β (Aβ) PET has emerged as clinically useful for more accurate diagnosis of patients with cognitive decline. Aβ deposition is a necessary cause or response to the cellular pathology of Alzheimer's disease (AD). Usual clinical and research interpretation of amyloid PET does not fully utilise all information regarding the spatial distribution of signal. We present a data-driven, spatially informed classifier to boost the diagnostic power of amyloid PET in AD.

METHODS

Voxel-wise k-means clustering of amyloid-positive voxels was performed; clusters were mapped to brain anatomy and tested for their associations by diagnostic category and disease severity with 758 amyloid PET scans from volunteers in the AD continuum from the Alzheimer's Disease Neuroimaging Initiative (ADNI). A machine learning approach based on this spatially constrained model using an optimised quadratic support vector machine was developed for automatic classification of scans for AD vs non-AD pathology.

RESULTS

This classifier boosted the accuracy of classification of AD scans to 81% using the amyloid PET alone with an area under the curve (AUC) of 0.91 compared to other spatial methods. This increased sensitivity to detect AD by 15% and the AUC by 9% compared to the use of a composite region of interest SUVr.

CONCLUSIONS

The diagnostic classification accuracy of amyloid PET was improved using an automated data-driven spatial classifier. Our classifier highlights the importance of considering the spatial variation in Aβ PET signal for optimal interpretation of scans. The algorithm now is available to be evaluated prospectively as a tool for automated clinical decision support in research settings.

Prediction of post-stroke cognitive impairment using brain FDG PET: deep learning-based approach

PURPOSE

Post-stroke cognitive impairment can affect up to one third of stroke survivors. Since cognitive function greatly contributes to patients' quality of life, an objective quantitative biomarker for early prediction of dementia after stroke is required. We developed a deep-learning (DL)-based signature using positron emission tomography (PET) to objectively evaluate cognitive decline in patients with stroke.

METHODS

We built a DL model that differentiated Alzheimer's disease (AD) from normal controls (NC) using brain fluorodeoxyglucose (FDG) PET from the Alzheimer's Disease Neuroimaging Initiative database. The model was directly transferred to a prospectively enrolled cohort of patients with stroke to differentiate patients with dementia from those without dementia. The accuracy of the model was evaluated by the area under the curve values of receiver operating characteristic curves (AUC-ROC). We visualized the distribution of DL-based features and brain regions that the model weighted for classification. Correlations between cognitive signature from the DL model and clinical variables were evaluated, and survival analysis for post-stroke dementia was performed in patients with stroke.

RESULTS

The classification of AD vs. NC subjects was performed with AUC-ROC of 0.94 (95% confidence interval [CI], 0.89-0.98). The transferred model discriminated stroke patients with dementia (AUC-ROC = 0.75). The score of cognitive decline signature using FDG PET was positively correlated with age, neutrophil-lymphocyte ratio and platelet-lymphocyte ratio and negatively correlated with body mass index in patients with stroke. We found that the cognitive decline score was an independent risk factor for dementia following stroke (hazard ratio, 10.90; 95% CI, 3.59-33.09; P < 0.0001) after adjustment for other key variables.

CONCLUSION

The DL-based cognitive signature using FDG PET was successfully transferred to an independent stroke cohort. It is suggested that DL-based cognitive evaluation using FDG PET could be utilized as an objective biomarker for cognitive dysfunction in patients with cerebrovascular diseases.

Data-driven respiratory phase-matched PET attenuation correction without CT

We propose a deep learning-based data-driven respiratory phase-matched gated-PET attenuation correction (AC) method that does not need a gated-CT. The proposed method is a multi-step process that consists of data-driven respiratory gating, gated attenuation map estimation using maximum-likelihood reconstruction of attenuation and activity (MLAA) algorithm, and enhancement of the gated attenuation maps using convolutional neural network (CNN). The gated MLAA attenuation maps enhanced by the CNN allowed for the phase-matched AC of gated-PET images. We conducted a non-rigid registration of the gated-PET images to generate motion-free PET images. We trained the CNN by conducting a 3D patch-based learning with 80 oncologic whole-body¹⁸F-fluorodeoxyglucose (¹⁸F-FDG) PET/CT scan data and applied it to seven regional PET/CT scans that cover the lower lung and upper liver. We investigated the impact of the proposed respiratory phase-matched AC of PET without utilizing CT on tumor size and standard uptake value (SUV) assessment, and PET image quality (%STD). The attenuation corrected gated and motion-free PET images generated using the proposed method yielded sharper organ boundaries and better noise characteristics than conventional gated and ungated PET images. A banana artifact observed in a phase-mismatched CT-based AC was not observed in the proposed approach. By employing the proposed method, the size of tumor was reduced by 12.3% and SUV_90%was increased by 13.3% in tumors with larger movements than 5 mm. %STD of liver uptake was reduced by 11.1%. The deep learning-based data-driven respiratory phase-matched AC method improved the PET image quality and reduced the motion artifacts.

Deep-learning-based cardiac amyloidosis classification from early acquired pet images

The objective of the present work was to evaluate the potential of deep learning tools for characterizing the presence of cardiac amyloidosis from early acquired PET images, i.e. 15 min after [18F]-Florbetaben tracer injection. 47 subjects were included in the study: 13 patients with transthyretin-related amyloidosis cardiac amyloidosis (ATTR-CA), 15 patients with immunoglobulin light-chain amyloidosis (AL-CA), and 19 control-patients (CTRL). [18F]-Florbetaben PET/CT images were acquired in list mode and data was sorted into a sinogram, covering a time interval of 5 min starting 15 min after the injection. The resulting sinogram was reconstructed using OSEM iterative algorithm. A deep convolutional neural network (CAclassNet) was designed and implemented, consisting of five 2D convolutional layers, three fully connected layers and a final classifier returning AL, ATTR and CTRL scores. A total of 1107 2D images (375 from AL-subtype patients, 312 from ATTR-subtype, and 420 from Controls) have been considered in the study and used to train, validate and test the proposed network. CAclassNet cross-validation resulted with train error mean ± sd of 2.001% ± 0.96%, validation error of 4.5% ± 2.26%, and net accuracy of 95.49% ± 2.26%. Network test error resulted in a mean ± sd values of 10.73% ± 0.76%. Sensitivity, specificity, and accuracy evaluated on the test dataset were respectively for AL-CA sub-type: 1, 0.912, 0.936; for ATTR-CA: 0.935, 0.897, 0.972; for control subjects: 0.809, 0.971, 0.909. In conclusion, the proposed CAclassNet model seems very promising as an aid for the clinician in the diagnosis of CA from cardiac [18F]-Florbetaben PET images acquired a few minutes after the injection.