Portable MRI and AI improve Alzheimer's disease diagnosis with cost-effective accuracy

Revolutionizing dementia care: See how a portable, AI-powered MRI system is breaking down barriers to Alzheimer's disease diagnosis, enabling early detection and global access.

study: Portable low-field magnetic resonance imaging for Alzheimer's disease assessment. Image credit: illustrissima / Shutterstock

recent nature communications In this study, we developed a machine learning pipeline to optimize portable LF-MRI acquisition and estimate brain morphometry and white matter hyperintensity (WMH) for Alzheimer's disease diagnosis.

Alzheimer's disease (AD): pathology and diagnosis

AD is a progressive neurodegenerative disease that affects memory, thinking, and behavior. Pathologically, it is characterized by amyloid-β (Aβ) deposition and the development of neurofibrillary tangles in the brain. Over time, increased accumulation of these proteins causes deleterious changes in brain structure and increases vascular damage. These are determined by quantifiable brain atrophy and WMH, respectively.

The progressive presymptomatic phase of AD typically lasts 10 to 20 years. This may be why 75% of dementia patients go undiagnosed for a long time. The availability of anti-amyloid therapies has increased the urgency of early detection in people with AD or mild cognitive impairment (MCI), as early diagnosis increases treatment efficacy.

Diagnosis of Alzheimer's disease is based on cognitive tests that assess Aβ and phosphotau burden using body fluid biomarkers, positron emission tomography (PET), and magnetic resonance imaging (MRI). Clinicians can determine changes in brain structure and integrity from multicontrast MRI. These imaging indicators include systemic atrophy and hippocampal atrophy and can help doctors understand the progression of the disease and the course of cognitive decline.

Neuroimaging is extremely useful in the diagnosis and management of Alzheimer's disease and MCI, but local and global inaccessibility contributes to underdiagnosis. Previous studies have highlighted the development of portable LF-MRI, which can effectively increase accessibility and potentially improve the diagnosis of various neurodegenerative diseases. This study highlighted the safety profile of LF-MRI and its potential as a low-cost point-of-care scan. However, as the magnetic field strength decreases, the signal-to-noise ratio (SNR) decreases, which affects image resolution.

About research

In the present study, we addressed the aforementioned limitations of LF-MRI in AD and MCI diagnosis by developing a machine learning tool that can automatically quantify brain morphometry and white matter lesions.

An imaging pipeline was established to help quantify brain volume. A sophisticated super-resolution and contrast synthesis technique (LF-SynthSR) was optimized to increase the resolution of LF images with subsequent segmentation (SynthSeg). For example, hippocampal volumes obtained from LF-MRI showed close agreement with their high-field MRI counterparts, with an absolute symmetrization percent difference (ASPD) of 2.8% and a Dice similarity coefficient of 0.87. This strategy helped establish optimal LF acquisition parameters for accurate quantification. This enabled white matter hyperintensity (WMH) burden (WMH-SynthSeg) measurements using automatic segmentation of WMH lesions from T2 fluid-attenuated inversion recovery (FLAIR) images acquired in the LF. This study validated LF-SynthSR, SynthSeg, and WMH-SynthSeg using a prospective cohort of patients diagnosed with MCI or AD.

To establish the imaging pipeline, three cohorts of participants were included and underwent MRI acquisition with portable low-field 0.064 T MRI and high-field conventional scans with field strengths of 1.5 to 3 T. The first cohort included 20 healthy individuals. Individuals (10 men, 10 women) with no history of neurological disease or memory impairment.

The second cohort included 23 participants (11 men and 12 women) with at least one vascular risk factor. However, none of the participants had a history of neurological complaints or memory problems. The third cohort included 54 individuals (32 men and 22 women) diagnosed with MCI or AD. These participants underwent an LF-MRI imaging protocol that included T1w, T2w, and FLAIR sequences.

Research results

Although the LF-MRI images did not have sufficient resolution for automated segmentation by high-field software analysis tools, they were initially Resolved (SR). -RAGE)-like image. In this study, we found that an isotropic voxel size of 3 mm or less improved segmentation accuracy and resulted in ASPD values ​​of less than 5% for hippocampal volume. Additionally, improvements to the LF-SynthSR v2 pipeline improve automatic segmentation accuracy and ease of use for low-field imaging applications.

In the first cohort, we demonstrated the effectiveness of automated segmentation by comparing AD-related segmentation volumes of the hippocampus, lateral ventricle, and whole brain generated from the original LF-SynthSR and LF-SynthSR v2 with conventional high-field (HF) MRI. Accuracy was evaluated. Obtained at 3T.

The accuracy of lateral ventricular volume was improved by comparing LF-SynthSR v2 and LF-SynthSR v1. Image acquisition times ranged from 1:53 to 9:48 min depending on voxel size and sequence. This study also found that an isotropic voxel size of 3 mm or less improves segmentation accuracy, especially in low SNR regions of LF-MRI. The accuracy of brain morphometry was found to be affected by voxel size and shape. Additionally, the LF-SynthSR v2 segmentation pipeline was validated against HF T1w MP-RAGE segmentation derived from the FreeSurfer segmentation tool ASEG.

WMH lesions due to axonal loss or cerebral small vessel disease are common in patients with cognitive impairment and were quantified using WMH-SynthSeg. Using these findings in FLAIR as T2 hyperintense lesions and automatically quantifying these lesions improved the AD diagnosis and monitoring ability of LF-MRI.

In this study, we used machine learning to generate WMH lesion volumes (WMHv) from LF-FLAIR images using WMH-SynthSeg. This strategy enabled simultaneous segmentation of WMH T2 FLAIR lesions in addition to previous brain morphometry. WMH volumes were strongly correlated with manual annotation and high-field imaging standards.

Based on the WMHv generated by WMH-SynthSeg, this machine learning tool was validated to be able to detect MCI, AD, and cognitively normal (CN) patients.

conclusion

The present study demonstrated that LF-MRI with machine learning tools can diagnose patients with AD or MCI. In the future, the device may also be evaluated for its ability to detect neurodegenerative tauopathies and vascular dementia. Its portability, low cost, and automated analysis pipeline suggest great potential to address global diagnostic disparities.