Staging and quantification of florbetaben PET images using machine learning: impact of predicted regional cortical tracer uptake and amyloid stage on clinical outcomes

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.

The Effects of PICALM rs3851179 and Age on Brain Atrophy and Cognition Along the Alzheimer's Disease Continuum

Rs3851179, a variant of PICALM gene, and age are the risk factors of Alzheimer's disease (AD). AD is divided into early-onset AD (EOAD, < 65 years) and late-onset AD (LOAD, ≥ 65 years) by age. The purpose was to investigate the impact of different genotypes of PICALM rs3851179 on brain atrophy and cognitive decline across the AD continuum in different age groups. Four hundred seven cognitive normal (CN) controls, 362 mild cognitive impairment (MCI) patients, and 94 AD patients were enrolled to assess the interaction between AD continuum, age status, and PICALM on gray matter volume (GMV), global cognition, memory function, and executive function using full factorial ANCOVA (3 × 2 × 2). The interaction between AD continuum and PICALM significantly affected the GMV of the left putamen (PUT.L). rs3851179 A-allele carriers did not show a significant decrease in PUT.L GMV from CN to MCI to AD, while GG-allele carriers did. The interaction between AD continuum and age status was significant on GMV of the left angular gyrus (ANG.L) and right superior occipital gyrus (SOG.R). LOAD had higher GMV of ANG.L and SOG.R than EOAD. The interactive effects among AD continuum, age status, and PICALM were not significant on GMV but were significant on global cognition and executive function. The A-allele was found to have a protective effect on global cognition and executive function in EOAD, but not significantly so in LOAD. PICALM rs3851179 A-allele might alleviate the atrophy of PUT.L across the AD continuum than GG-allele. Age status did not affect the interaction between AD continuum and PICALM on brain atrophy. The ANG.L and SOG.R atrophied more severely in EOAD than in LOAD. Rs3851179 A-allele was protective for global cognition and executive function in EOAD.

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.

Validation of quantitative assessment of florbetaben PET scans as an adjunct to the visual assessment across 15 software methods

PURPOSE

Amyloid positron emission tomography (PET) with [18F]florbetaben (FBB) is an established tool for detecting Aβ deposition in the brain in vivo based on visual assessment of PET scans. Quantitative measures are commonly used in the research context and allow continuous measurement of amyloid burden. The aim of this study was to demonstrate the robustness of FBB PET quantification.

METHODS

This is a retrospective analysis of FBB PET images from 589 subjects. PET scans were quantified with 15 analytical methods using nine software packages (MIMneuro, Hermes BRASS, Neurocloud, Neurology Toolkit, statistical parametric mapping (SPM8), PMOD Neuro, CapAIBL, non-negative matrix factorization (NMF), AmyloidIQ) that used several metrics to estimate Aβ load (SUVR, centiloid, amyloid load, and amyloid index). Six analytical methods reported centiloid (MIMneuro, standard centiloid, Neurology Toolkit, SPM8 (PET only), CapAIBL, NMF). All results were quality controlled.

RESULTS

The mean sensitivity, specificity, and accuracy were 96.1 ± 1.6%, 96.9 ± 1.0%, and 96.4 ± 1.1%, respectively, for all quantitative methods tested when compared to histopathology, where available. The mean percentage of agreement between binary quantitative assessment across all 15 methods and visual majority assessment was 92.4 ± 1.5%. Assessments of reliability, correlation analyses, and comparisons across software packages showed excellent performance and consistent results between analytical methods.

CONCLUSION

This study demonstrated that quantitative methods using both CE marked software and other widely available processing tools provided comparable results to visual assessments of FBB PET scans. Software quantification methods, such as centiloid analysis, can complement visual assessment of FBB PET images and could be used in the future for identification of early amyloid deposition, monitoring disease progression and treatment effectiveness.

International Nuclear Medicine Consensus on the Clinical Use of Amyloid Positron Emission Tomography in Alzheimer's Disease

Alzheimer's disease (AD) is the main cause of dementia, with its diagnosis and management remaining challenging. Amyloid positron emission tomography (PET) has become increasingly important in medical practice for patients with AD. To integrate and update previous guidelines in the field, a task group of experts of several disciplines from multiple countries was assembled, and they revised and approved the content related to the application of amyloid PET in the medical settings of cognitively impaired individuals, focusing on clinical scenarios, patient preparation, administered activities, as well as image acquisition, processing, interpretation and reporting. In addition, expert opinions, practices, and protocols of prominent research institutions performing research on amyloid PET of dementia are integrated. With the increasing availability of amyloid PET imaging, a complete and standard pipeline for the entire examination process is essential for clinical practice. This international consensus and practice guideline will help to promote proper clinical use of amyloid PET imaging in patients with AD.

Comparison of Three Automated Approaches for Classification of Amyloid-PET Images

Automated amyloid-PET image classification can support clinical assessment and increase diagnostic confidence. Three automated approaches using global cut-points derived from Receiver Operating Characteristic (ROC) analysis, machine learning (ML) algorithms with regional SUVr values, and deep learning (DL) network with 3D image input were compared under various conditions: number of training data, radiotracers, and cohorts. 276 [¹¹C]PiB and 209 [¹⁸F]AV45 PET images from ADNI database and our local cohort were used. Global mean and maximum SUVr cut-points were derived using ROC analysis. 68 ML models were built using regional SUVr values and one DL network was trained with classifications of two visual assessments - manufacturer's recommendations (gray-scale) and with visually guided reference region scaling (rainbow-scale). ML-based classification achieved similarly high accuracy as ROC classification, but had better convergence between training and unseen data, with a smaller number of training data. Naïve Bayes performed the best overall among the 68 ML algorithms. Classification with maximum SUVr cut-points yielded higher accuracy than with mean SUVr cut-points, particularly for cohorts showing more focal uptake. DL networks can support the classification of definite cases accurately but performed poorly for equivocal cases. Rainbow-scale standardized image intensity scaling and improved inter-rater agreement. Gray-scale detects focal accumulation better, thus classifying more amyloid-positive scans. All three approaches generally achieved higher accuracy when trained with rainbow-scale classification. ML yielded similarly high accuracy as ROC, but with better convergence between training and unseen data, and further work may lead to even more accurate ML methods.

Classification of Alzheimer’s Disease Using Dual-Phase 18F-Florbetaben Image with Rank-Based Feature Selection and Machine Learning

18F-florbetaben (FBB) positron emission tomography is a representative imaging test that observes amyloid deposition in the brain. Compared to delay-phase FBB (dFBB), early-phase FBB shows patterns related to glucose metabolism in 18F-fluorodeoxyglucose perfusion images. The purpose of this study is to prove that classification accuracy is higher when using dual-phase FBB (dual FBB) versus dFBB quantitative analysis by using machine learning and to find an optimal machine learning model suitable for dual FBB quantitative analysis data. The key features of our method are (1) a feature ranking method for each phase of FBB with a cross-validated F1 score and (2) a quantitative diagnostic model based on machine learning methods. We compared four classification models: support vector machine, naïve Bayes, logistic regression, and random forest (RF). In composite standardized uptake value ratio, RF achieved the best performance (F1: 78.06%) with dual FBB, which was 4.83% higher than the result with dFBB. In conclusion, regardless of the two quantitative analysis methods, using the dual FBB has a higher classification accuracy than using the dFBB. The RF model is the machine learning model that best classifies a dual FBB. The regions that have the greatest influence on the classification of dual FBB are the frontal and temporal lobes.

The impact of subthreshold levels of amyloid deposition on conversion to dementia in patients with amyloid-negative amnestic mild cognitive impairment

BACKGROUND

About 40-50% of patients with amnestic mild cognitive impairment (MCI) are found to have no significant Alzheimer's pathology based on amyloid PET positivity. Notably, conversion to dementia in this population is known to occur much less often than in amyloid-positive MCI. However, the relationship between MCI and brain amyloid deposition remains largely unknown. Therefore, we investigated the influence of subthreshold levels of amyloid deposition on conversion to dementia in amnestic MCI patients with negative amyloid PET scans.

METHODS

This study was a retrospective cohort study of patients with amyloid-negative amnestic MCI who visited the memory clinic of Asan Medical Center. All participants underwent detailed neuropsychological testing, brain magnetic resonance imaging, and [18F]-florbetaben (FBB) positron emission tomography scan (PET). Conversion to dementia was determined by a neurologist based on a clinical interview with a detailed neuropsychological test or a decline in the Korean version of the Mini-Mental State Examination score of more than 4 points per year combined with impaired activities of daily living. Regional cortical amyloid levels were calculated, and a receiver operating characteristic (ROC) curve for conversion to dementia was obtained. To increase the reliability of the results of the study, we analyzed the Alzheimer's Disease Neuroimaging Initiative (ADNI) dataset together.

RESULTS

During the follow-up period, 36% (39/107) of patients converted to dementia from amnestic MCI. The dementia converter group displayed increased standardized uptake value ratio (SUVR) values of FBB on PET in the bilateral temporal, parietal, posterior cingulate, occipital, and left precuneus cortices as well as increased global SUVR. Among volume of interests, the left parietal SUVR predicted conversion to dementia with the highest accuracy in the ROC analysis (area under the curve [AUC] = 0.762, P < 0.001). The combination of precuneus, parietal cortex, and FBB composite SUVRs also showed a higher accuracy in predicting conversion to dementia than other models (AUC = 0.763). Of the results of ADNI data, the SUVR of the left precuneus SUVR showed the highest AUC (AUC = 0.596, P = 0.006).

CONCLUSION

Our findings suggest that subthreshold amyloid levels may contribute to conversion to dementia in patients with amyloid-negative amnestic MCI.

Role of White Matter Abnormalities in the Relationship Between Microbleed Burden and Cognitive Impairment in Cerebral Amyloid Angiopathy

Background: Cerebral amyloid angiopathy (CAA) often presents as cognitive impairment, but the mechanism of cognitive decline is unclear. Recent studies showed that number of microbleeds were associated with cognitive decline. Objective: We aimed to investigate how microbleeds contribute to cognitive impairment in association with white matter tract abnormalities or cortical thickness in CAA. Methods: This retrospective comparative study involved patients with probable CAA according to the Boston criteria (Aβ  + CAA) and patients with Alzheimer’s disease (Aβ  + AD), all of whom showed severe amyloid deposition on amyloid PET. Using mediation analysis, we investigated how FA or cortical thickness mediates the correlation between the number of lobar microbleeds and cognition. Results: We analyzed 30 patients with Aβ  + CAA (age 72.2±7.6, female 53.3%) and 30 patients with Aβ  + AD (age 71.5±7.6, female 53.3%). The two groups showed similar degrees of cortical amyloid deposition in AD-related regions. The Aβ  + CAA group had significantly lower FA values in the clusters of the posterior area than did the Aβ  + AD group (family-wise error-corrected p <  0.05). The correlation between the number of lobar microbleeds and visuospatial function was indirectly mediated by white matter tract abnormality of right posterior thalamic radiation (PTR) and tapetum, while lobar microbleeds and language function was indirectly mediated by the abnormality of left PTR and sagittal stratum. Cortical thickness did not mediate the association between lobar microbleeds and cognition. Conclusion: This result supports the hypothesis that microbleeds burden leads to white matter tract damage and subsequent cognitive decline in CAA.

Harmonisation of PET imaging features with different amyloid ligands using machine learning-based classifier

PURPOSE

In this study, we used machine learning to develop a new method derived from a ligand-independent amyloid (Aβ) positron emission tomography (PET) classifier to harmonise different Aβ ligands.

METHODS

We obtained 107 paired 18F-florbetaben (FBB) and 18F-flutemetamol (FMM) PET images at the Samsung Medical Centre. To apply the method to FMM ligand, we transferred the previously developed FBB PET classifier to test similar features from the FMM PET images for application to FMM, which in turn developed a ligand-independent Aβ PET classifier. We explored the concordance rates of our classifier in detecting cortical and striatal Aβ positivity. We investigated the correlation of machine learning-based cortical tracer uptake (ML-CTU) values quantified by the classifier between FBB and FMM.

RESULTS

This classifier achieved high classification accuracy (area under the curve = 0.958) even with different Aβ PET ligands. In addition, the concordance rate of FBB and FMM using the classifier (87.5%) was good to excellent, which seemed to be higher than that in visual assessment (82.7%) and lower than that in standardised uptake value ratio cut-off categorisation (93.3%). FBB and FMM ML-CTU values were highly correlated with each other (R = 0.903).

CONCLUSION

Our findings suggested that our novel classifier may harmonise FBB and FMM ligands in the clinical setting which in turn facilitate the biomarker-guided diagnosis and trials of anti-Aβ treatment in the research field.

Head-to-Head Comparison of 18F-Florbetaben and 18F-Flutemetamol in the Cortical and Striatal Regions

BACKGROUND

18F-florbetaben (FBB) and 18F-flutemetamol (FMM) amyloid PET have been developed and approved for clinical use. It is important to understand the distinct features of these ligands to compare and correctly interpret the results of different amyloid PET studies.

OBJECTIVE

We performed a head-to-head comparison of FBB and FMM to compare with regard to imaging characteristics, including dynamic range of retention, and differences in quantitative measurements between the two ligands in cortical, striatal, and white matter (WM) regions.

METHODS

Paired FBB and FMM PET images were acquired in 107 participants. Correlations of FBB and FMM amyloid deposition in the cortex, striatum, and WM were investigated and compared in different reference regions (cerebellar gray matter (CG), whole cerebellum (WC), WC with brainstem (WC + B), and pons).

RESULTS

The cortical SUVR (R2 = 0.97) and striatal SUVR (R2 = 0.95) demonstrated an excellent linear correlation between FBB and FMM using a WC as reference region. There was no difference in the cortical SUVR ratio between the two ligands (p = 0.90), but the striatal SUVR ratio was higher in FMM than in FBB (p < 0.001). Also, the effect size of differences in striatal SUVR seemed to be higher with FMM (2.61) than with FBB (2.34). These trends were similarly observed according to four different reference regions (CG, WC, WC + B, and pons).

CONCLUSION

Our findings suggest that FMM might be better than FBB to detect amyloid burden in the striatum, although both ligands are comparable for imaging AD pathology in vivo.

A review of the application of machine learning in molecular imaging

Molecular imaging (MI) is a science that uses imaging methods to reflect the changes of molecular level in living state and conduct qualitative and quantitative studies on its biological behaviors in imaging. Optical molecular imaging (OMI) and nuclear medical imaging are two key research fields of MI. OMI technology refers to the optical information generated by the imaging target (such as tumors) due to drug intervention and other reasons. By collecting the optical information, researchers can track the motion trajectory of the imaging target at the molecular level. Owing to its high specificity and sensitivity, OMI has been widely used in preclinical research and clinical surgery. Nuclear medical imaging mainly detects ionizing radiation emitted by radioactive substances. It can provide molecular information for early diagnosis, effective treatment and basic research of diseases, which has become one of the frontiers and hot topics in the field of medicine in the world today. Both OMI and nuclear medical imaging technology require a lot of data processing and analysis. In recent years, artificial intelligence technology, especially neural network-based machine learning (ML) technology, has been widely used in MI because of its powerful data processing capability. It provides a feasible strategy to deal with large and complex data for the requirement of MI. In this review, we will focus on the applications of ML methods in OMI and nuclear medical imaging.

A multisite analysis of the concordance between visual image interpretation and quantitative analysis of [18F]flutemetamol amyloid PET images

BACKGROUND

[¹⁸F]flutemetamol PET scanning provides information on brain amyloid load and has been approved for routine clinical use based upon visual interpretation as either negative (equating to none or sparse amyloid plaques) or amyloid positive (equating to moderate or frequent plaques). Quantitation is however fundamental to the practice of nuclear medicine and hence can be used to supplement amyloid reading methodology especially in unclear cases.

METHODS

A total of 2770 [¹⁸F]flutemetamol images were collected from 3 clinical studies and 6 research cohorts with available visual reading of [¹⁸F]flutemetamol and quantitative analysis of images. These were assessed further to examine both the discordance and concordance between visual and quantitative imaging primarily using thresholds robustly established using pathology as the standard of truth. Scans covered a wide range of cases (i.e. from cognitively unimpaired subjects to patients attending the memory clinics). Methods of quantifying amyloid ranged from using CE/510K cleared marked software (e.g. CortexID, Brass), to other research-based methods (e.g. PMOD, CapAIBL). Additionally, the clinical follow-up of two types of discordance between visual and quantitation (V+Q- and V-Q+) was examined with competing risk regression analysis to assess possible differences in prediction for progression to Alzheimer's disease (AD) and other diagnoses (OD).

RESULTS

Weighted mean concordance between visual and quantitation using the autopsy-derived threshold was 94% using pons as the reference region. Concordance from a sensitivity analysis which assessed the maximum agreement for each cohort using a range of cut-off values was also estimated at approximately 96% (weighted mean). Agreement was generally higher in clinical cases compared to research cases. V-Q+ discordant cases were 11% more likely to progress to AD than V+Q- for the SUVr with pons as reference region.

CONCLUSIONS

Quantitation of amyloid PET shows a high agreement vs binary visual reading and also allows for a continuous measure that, in conjunction with possible discordant analysis, could be used in the future to identify possible earlier pathological deposition as well as monitor disease progression and treatment effectiveness.

Machine Learning in Nuclear Medicine: Part 2-Neural Networks and Clinical Aspects

This article is the second part in our machine learning series. Part 1 provided a general overview of machine learning in nuclear medicine. Part 2 focuses on neural networks. We start with an example illustrating how neural networks work and a discussion of potential applications. Recognizing that there is a spectrum of applications, we focus on recent publications in the areas of image reconstruction, low-dose PET, disease detection, and models used for diagnosis and outcome prediction. Finally, since the way machine learning algorithms are reported in the literature is extremely variable, we conclude with a call to arms regarding the need for standardized reporting of design and outcome metrics and we propose a basic checklist our community might follow going forward.