Classification of amyloid status using machine learning with histograms of oriented 3D gradients

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.

A Review on the Use of Modern Computational Methods in Alzheimer's Disease-Detection and Prediction

Discoveries in the field of medical sciences are blooming rapidly at the cost of voluminous efforts. Presently, multidisciplinary research activities have been especially contributing to catering cutting-edge solutions to critical problems in the domain of medical sciences. The modern age computing resources have proved to be a boon in this context. Effortless solutions have become a reality, and thus, the real beneficiary patients are able to enjoy improved lives. One of the most emerging problems in this context is Alzheimer's disease, an incurable neurological disorder. For this, early diagnosis is made possible with benchmark computing tools and schemes. These benchmark schemes are the results of novel research contributions being made intermittently in the timeline. In this review, an attempt is made to explore all such contributions in the past few decades. A systematic review is made by categorizing these contributions into three folds, namely, First, Second, and Third Generations. However, priority is given to the latest ones as a handful of literature reviews are already available for the classical ones. Key contributions are discussed vividly. The objectives set for this review are to bring forth the latest discoveries in computing methodologies, especially those dedicated to the diagnosis of Alzheimer's disease. A detailed timeline of the contributions is also made available. Performance plots for certain key contributions are also presented for better graphical understanding.

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.

Invariance encoding in sliced-Wasserstein space for image classification with limited training data

Deep convolutional neural networks (CNNs) are broadly considered to be state-of-the-art generic end-to-end image classification systems. However, they are known to underperform when training data are limited and thus require data augmentation strategies that render the method computationally expensive and not always effective. Rather than using a data augmentation strategy to encode invariances as typically done in machine learning, here we propose to mathematically augment a nearest subspace classification model in sliced-Wasserstein space by exploiting certain mathematical properties of the Radon Cumulative Distribution Transform (R-CDT), a recently introduced image transform. We demonstrate that for a particular type of learning problem, our mathematical solution has advantages over data augmentation with deep CNNs in terms of classification accuracy and computational complexity, and is particularly effective under a limited training data setting. The method is simple, effective, computationally efficient, non-iterative, and requires no parameters to be tuned. Python code implementing our method is available at https://github.com/rohdelab/mathematical augmentation. Our method is integrated as a part of the software package PyTransKit, which is available at https://github.com/rohdelab/PyTransKit.

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.

The Value of Neuroimaging in Dementia Diagnosis

PURPOSE OF REVIEW

This article discusses neuroimaging in dementia diagnosis, with a focus on new applications of MRI and positron emission tomography (PET).

RECENT FINDINGS

Although the historical use of MRI in dementia diagnosis has been supportive to exclude structural etiologies, recent innovations allow for quantification of atrophy patterns that improve sensitivity for supporting the diagnosis of dementia causes. Neuronuclear approaches allow for localization of specific amyloid and tau neuropathology on PET and are available for clinical use, in addition to dopamine transporter scans in dementia with Lewy bodies and metabolic studies with fludeoxyglucose PET (FDG-PET).

SUMMARY

Using computerized software programs for MRI analysis and cross-sectional and longitudinal evaluations of hippocampal, ventricular, and lobar volumes improves sensitivity in support of the diagnosis of Alzheimer disease and frontotemporal dementia. MRI protocol requirements for such quantification are three-dimensional T1-weighted volumetric imaging protocols, which may need to be specifically requested. Fluid-attenuated inversion recovery (FLAIR) and 3.0T susceptibility-weighted imaging (SWI) sequences are useful for the detection of white matter hyperintensities as well as microhemorrhages in vascular dementia and cerebral amyloid angiopathy. PET studies for amyloid and/or tau pathology can add additional specificity to the diagnosis but currently remain largely inaccessible outside of research settings because of prohibitive cost constraints in most of the world. Dopamine transporter PET scans can help identify Lewy body dementia and are thus of potential clinical value.

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.

Multiparametric magnetic resonance imaging and positron emission tomography findings in neurodegenerative diseases: Current status and future directions

Neurodegenerative diseases (NDDs) are characterized by progressive neuronal loss, leading to dementia and movement disorders. NDDs broadly include Alzheimer's disease, frontotemporal lobar degeneration, parkinsonian syndromes, and prion diseases. There is an ever-increasing prevalence of mild cognitive impairment and dementia, with an accompanying immense economic impact, prompting efforts aimed at early identification and effective interventions. Neuroimaging is an essential tool for the early diagnosis of NDDs in both clinical and research settings. Structural, functional, and metabolic imaging modalities, including magnetic resonance imaging (MRI) and positron emission tomography (PET), are widely available. They show encouraging results for diagnosis, monitoring, and treatment response evaluation. The current review focuses on the complementary role of various imaging modalities in relation to NDDs, the qualitative and quantitative utility of newer MRI techniques, novel radiopharmaceuticals, and integrated PET/MRI in the setting of NDDs.

Application of Deep Learning to Predict Standardized Uptake Value Ratio and Amyloid Status on 18F-Florbetapir PET Using ADNI Data

BACKGROUND AND PURPOSE

Cortical amyloid quantification on PET by using the standardized uptake value ratio is valuable for research studies and clinical trials in Alzheimer disease. However, it is resource intensive, requiring co-registered MR imaging data and specialized segmentation software. We investigated the use of deep learning to automatically quantify standardized uptake value ratio and used this for classification.

MATERIALS AND METHODS

Using the Alzheimer's Disease Neuroimaging Initiative dataset, we identified 2582 18F-florbetapir PET scans, which were separated into positive and negative cases by using a standardized uptake value ratio threshold of 1.1. We trained convolutional neural networks (ResNet-50 and ResNet-152) to predict standardized uptake value ratio and classify amyloid status. We assessed performance based on network depth, number of PET input slices, and use of ImageNet pretraining. We also assessed human performance with 3 readers in a subset of 100 randomly selected cases.

RESULTS

We have found that 48% of cases were amyloid positive. The best performance was seen for ResNet-50 by using regression before classification, 3 input PET slices, and pretraining, with a standardized uptake value ratio root-mean-square error of 0.054, corresponding to 95.1% correct amyloid status prediction. Using more than 3 slices did not improve performance, but ImageNet initialization did. The best trained network was more accurate than humans (96% versus a mean of 88%, respectively).

CONCLUSIONS

Deep learning algorithms can estimate standardized uptake value ratio and use this to classify 18F-florbetapir PET scans. Such methods have promise to automate this laborious calculation, enabling quantitative measurements rapidly and in settings without extensive image processing manpower and expertise.

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

Purpose

We developed a machine learning–based classifier for in vivo amyloid positron emission tomography (PET) staging, quantified cortical uptake of the PET tracer by using a machine learning method, and investigated the impact of these amyloid PET parameters on clinical and structural outcomes.

Methods

A total of 337 ¹⁸F-florbetaben PET scans obtained at Samsung Medical Center were assessed. We defined a feature vector representing the change in PET tracer uptake from grey to white matter. Using support vector machine (SVM) regression and SVM classification, we quantified the cortical uptake as predicted regional cortical tracer uptake (pRCTU) and categorised the scans as positive and negative. Positive scans were further classified into two stages according to the striatal uptake. We compared outcome parameters among stages and further assessed the association between the pRCTU and outcome variables. Finally, we performed path analysis to determine mediation effects between PET variables.

Results

The classification accuracy was 97.3% for cortical amyloid positivity and 91.1% for striatal positivity. The left frontal and precuneus/posterior cingulate regions, as well as the anterior portion of the striatum, were important in determination of stages. The clinical scores and magnetic resonance imaging parameters showed negative associations with PET stage. However, except for the hippocampal volume, most outcomes were associated with the stage through the complete mediation effect of pRCTU.

Conclusion

Using a machine learning algorithm, we achieved high accuracy for in vivo amyloid PET staging. The in vivo amyloid stage was associated with cognitive function and cerebral atrophy mostly through the mediation effect of cortical amyloid.

Electronic supplementary material

The online version of this article (10.1007/s00259-019-04663-3) contains supplementary material, which is available to authorized users.

The clinical feasibility of deep learning-based classification of amyloid PET images in visually equivocal cases

PURPOSE

Although most deep learning (DL) studies have reported excellent classification accuracy, these studies usually target typical Alzheimer's disease (AD) and normal cognition (NC) for which conventional visual assessment performs well. A clinically relevant issue is the selection of high-risk subjects who need active surveillance among equivocal cases. We validated the clinical feasibility of DL compared with visual rating or quantitative measurement for assessing the diagnosis and prognosis of subjects with equivocal amyloid scans.

METHODS

¹⁸F-florbetaben scans of 430 cases (85 NC, 233 mild cognitive impairment, and 112 AD) were assessed through visual rating-based, quantification-based, and DL-based methods. DL was trained using 280 two-dimensional PET images (80%) and tested by randomly assigning the remaining (70 cases, 20%) cases and a clinical validation set of 54 equivocal cases. In the equivocal cases, we assessed the agreement among the visual rating, quantification, and DL and compared the clinical outcome according to each modality-based amyloid status.

RESULTS

The visual reading was positive in 175 cases, equivocal in 54 cases, and negative in 201 cases. The composite SUVR cutoff value was 1.32 (AUC 0.99). The subject-level performance of DL using the test set was 100%. Among the 54 equivocal cases, 37 cases were classified as positive (Eq(deep+)) by DL, 40 cases were classified by a second-round visual assessment, and 40 cases were classified by quantification. The DL- and quantification-based classifications showed good agreement (83%, κ = 0.59). The composite SUVRs differed between Eq(deep+) (1.47 [0.13]) and Eq(deep-) (1.29 [0.10]; P < 0.001). DL, but not the visual rating, showed a significant difference in the Mini-Mental Status Examination score change during the follow-up between Eq(deep+) (- 4.21 [0.57]) and Eq(deep-) (- 1.74 [0.76]; P = 0.023) (mean duration, 1.76 years).

CONCLUSIONS

In visually equivocal scans, DL was more related to quantification than to visual assessment, and the negative cases selected by DL showed no decline in cognitive outcome. DL is useful for clinical diagnosis and prognosis assessment in subjects with visually equivocal amyloid scans.

Structurally-sensitive Multi-scale Deep Neural Network for Low-Dose CT Denoising

Computed tomography (CT) is a popular medical imaging modality and enjoys wide clinical applications. At the same time, the x-ray radiation dose associated with CT scannings raises a public concern due to its potential risks to the patients. Over the past years, major efforts have been dedicated to the development of Low-Dose CT (LDCT) methods. However, the radiation dose reduction compromises the signal-to-noise ratio (SNR), leading to strong noise and artifacts that downgrade CT image quality. In this paper, we propose a novel 3D noise reduction method, called Structurally-sensitive Multi-scale Generative Adversarial Net (SMGAN), to improve the LDCT image quality. Specifically, we incorporate three-dimensional (3D) volumetric information to improve the image quality. Also, different loss functions for training denoising models are investigated. Experiments show that the proposed method can effectively preserve structural and textural information in reference to normal-dose CT (NDCT) images, and significantly suppress noise and artifacts. Qualitative visual assessments by three experienced radiologists demonstrate that the proposed method retrieves more information, and outperforms competing methods.

Optimized classification of 18F-Florbetaben PET scans as positive and negative using an SUVR quantitative approach and comparison to visual assessment

Introduction

Standardized uptake value ratios (SUVRs) calculated from cerebral cortical areas can be used to categorize ¹⁸F-Florbetaben (FBB) PET scans by applying appropriate cutoffs. The objective of this work was first to generate FBB SUVR cutoffs using visual assessment (VA) as standard of truth (SoT) for a number of reference regions (RR) (cerebellar gray matter (GCER), whole cerebellum (WCER), pons (PONS), and subcortical white matter (SWM)). Secondly, to validate the FBB PET scan categorization performed by SUVR cutoffs against the categorization made by post-mortem histopathological confirmation of the Aβ presence. Finally, to evaluate the added value of SUVR cutoff categorization to VA.

Methods

SUVR cutoffs were generated for each RR using FBB scans from 143 subjects who were visually assessed by 3 readers. SUVR cutoffs were validated in 78 end-of life subjects using VA from 8 independent blinded readers (3 expert readers and 5 non-expert readers) and histopathological confirmation of the presence of neuritic beta-amyloid plaques as SoT. Finally, the number of correctly or incorrectly classified scans according to pathology results using VA and SUVR cutoffs was compared.

Results

Composite SUVR cutoffs generated were 1.43 (GCER), 0.96 (WCER), 0.78 (PONS) and 0.71 (SWM). Accuracy values were high and consistent across RR (range 83–94% for histopathology, and 85–94% for VA). SUVR cutoff performed similarly as VA but did not improve VA classification of FBB scans read either by expert readers or the majority read but provided higher accuracy than some non-expert readers.

Conclusion

The accurate scan classification obtained in this study supports the use of VA as SoT to generate site-specific SUVR cutoffs. For an elderly end of life population, VA and SUVR cutoff categorization perform similarly in classifying FBB scans as Aβ-positive or Aβ-negative. These results emphasize the additional contribution that SUVR cutoff classification may have compared with VA performed by non-expert readers.

•

SUVR cutoffs to classify Florbetaben PET scans as positive and negative were generated.

•

SUVR cutoffs were validated against post-mortem histopathological confirmation.

•

Added value of SUVR cutoff classification to visual assessment was evaluated.

•

SUVR cutoff classification provided higher accuracy than some non-expert readers.

•

Results emphasize the contribution that SUVR cutoffs may have to visual assessment.