Our Institutional Review Board (IRB) approved this cross-sectional prospective study and informed consent was obtained from the participants who were recruited in the neurological center of our institution. Participants provided a detailed medical history and underwent a neurologic examination, standard neuropsychological testing, and MRI scan. The brain MR images of each participant were evaluated by neuroradiologists with more than 15 years of imaging experience to determine any evidence of prior cortical infarctions or other space-occupying lesions.

Cognitive function was assessed using Mini-Mental State Examination (MMSE), and based on the results of the neuropsychological examination and MRI findings, the patients with mild and probable AD were included in this study, defined as those having clinical dementia rating (CDR) scores of 0.5, 1, or 2 according to the criteria of the National Institute of Neurological and Communicative Disorders and the Stroke-Alzheimer's Disease and Related Disorders Association (20, 21). In addition, participants with amnestic MCI, either single-domain or multiple-domain, were also included according to the Petersen criteria (1, 22). Finally, elderly CN participants were selected from healthy volunteers who did not have a medical history of neurological disease, who showed normal results on detailed examination according to the Korean normative database, and who also had a normal brain MRI.

A total of 85 participants were enrolled in this study. Eleven participants were excluded from the subsequent analysis due to the following reasons: two participants had multiple infarctions, one participant had a brain tumor, one participant had a small traumatic subarachnoid hemorrhage lesion, two participants had severe hydrocephalus, one participant had a venous anomaly lesion, two participants had non-amnestic MCI, and two participants had metallic artifacts in the images. Therefore, the 74 remaining participants were grouped as 24 CN, 27 MCI, and 23 AD. Table 1 summarizes the demographic characteristics and the neuropsychological tests of the participant groups. There were several comorbid conditions of the participants in this study. A total of 33 subjects (44.6%) had one or more underlying health conditions and the most common comorbidity was hypertension (29.7%) in all groups. In the AD group with 23 participants, 7 subjects (30.4%) had hypertension and other comorbidities included in diabetes mellitus (2), unruptured intracranial cerebral aneurysm (1), chronic kidney disease (1), myelodysplastic syndrome (1), and transplanted liver (1). In the MCI group with 27 participants, 10 subjects (37%) had hypertension that is also the most common comorbid condition and other comorbidities were diabetes mellitus (5), unruptured aneurysm of the internal carotid artery (1), and ankylosing spondylitis (1). In the normal control group with 24 participants, 5 subjects (20.8%) had hypertension and 5 subjects had diabetes mellitus that was followed by chronic kidney disease (1), nephrotic syndrome (1), atrial fibrillation (1), dilated cardiomyopathy (1), and unruptured aneurysm of the internal carotid artery (1) as underlying conditions.

MRI was performed using a 3.0 Tesla MRI system equipped with a 32-channel sensitivity encoding head coil (Ingenia, Philips Medical System, Best, The Netherlands). For the brain MREPT images, a multi-echo turbo spin-echo pulse sequence was used (23). Before the data acquisition, we applied a volume shimming method with the volume defined to cover the brain region. The constant level appearance technique was used to achieve homogeneity correction by using coil sensitivity maps acquired in a reference scan. The imaging parameters were as follows: repetition time (TR) = 3,200 ms, first echo time (TE) = 12 ms with 12 ms intervals, flip angle (FA) = 90°, number of echoes (NE) = 6, number of average (NSA) = 1, slice thickness = 5 mm, number of slices = 20 without a gap between the slices, slice orientation = transverse, fold-over direction = anterior-posterior (AP), fat shift direction = left, acquired voxel size = 2 mm³ × 2 mm³ × 5 mm³, reconstructed voxel size = 1 mm³ × 1 mm³ × 5 mm³, acquisition matrix = 112 × 112, reconstruction matrix = 224 × 224, field-of-view (FOV) = 224 mm² × 224 mm², SENSE factor = 0, TSE factor = 6, RF shim = “adaptive,” B0 shim = “PB-volume,” slice scan order = interleaved, and regional saturation slab = 45 mm at the feet direction. The scan time of the MREPT sequence was 6 min and 5 s. Real and imaginary images were saved for reconstructing the conductivity map. For image registration and brain tissue segmentation, a sagittal structural three-dimensional (3D) T1-weighted (T1W) image was acquired with the fast field-echo (FFE) sequence, which is similar to the magnetization-prepared rapid acquisition of the gradient echo (MPRAGE) sequence. The imaging parameters were as follows: TR = 8.1 ms, TE = 3.7 ms, FA = 8°, FOV = 236 mm² × 236 mm², and voxel size = 1 mm³ × 1 mm³ × 1 mm³. T2-weighted turbo-spin-echo, fluid-attenuated inversion recovery (FLAIR), and gradient-echo images were also acquired to evaluate any brain abnormalities.

A home-made software was used to map the HFC at the Larmor frequency of 128 MHz at 3T (14). The MREPT formula based on a convection reaction equation was derived by adding the regularization coefficient (24). To solve the convection reaction partial differential equation, we used the 2-dimensional finite-difference method. Detailed information for reconstructing the HFC map was described in the reconstruction part in the Supplementary Material.

To evaluate HFC maps among the three participant groups, the post-processing was performed using the Statistical Parametric Mapping version 12 (SPM12) software (http://www.fil.ion.ucl.ac.uk/spm/software/spm12/). First, the 3D T1W image and the real MREPT image for each participant were co-registered. Second, the 3D T1W image was spatially normalized into our AD-specific brain template and segmented into gray matter and white matter using the computational anatomy toolbox (CAT12) tool (http://www.neuro.uni-jena.de/cat/). Third, the HFC map was then spatially normalized into the brain template using the deformation field information of the 3D T1W image. Finally, the Gaussian smoothing using the full-width at half maximum (FWHM) of 10 mm³ × 10 mm³ × 10 mm³ was applied to the HFC map as well as brain tissue volumes for voxel-based statistical analyses.

The ANOVA test showed that the age was not significantly different between the participant groups (F = 1.72, p = 0.19). The Kruskal–Wallis test showed that the MMSE scores (F = 61.6, p < 0.001) were significantly different between AD and other groups, as expected. The sex was not significantly different between the participant groups (χ² <1.404, p > 0.236). The results of the statistical analysis of the participants' demographic data and the neuropsychological test are summarized in Table 1.

Figure 1 shows the representative HFC maps with segmented brain tissues obtained from one CN participant (72-year-old female), one mild cognitive impairment (MCI) patient (74-year-old female), and one AD patient (73-year-old female). The HFC signal exhibits increased HFC in the AD patient compared with the CN participant and the MCI patient at the bilateral peri-insular temporal lobe and white matter near the lateral ventricle in the temporal lobe. In general, the HFC signal was high around the parietal lobe, insular, and cerebrospinal fluid (CSF) areas.

We obtained a good HFC map non-invasively and found that the HFC values were significantly increased with increasing disease severity (Table 2). Furthermore, the HFC value in the insula has a high AUC value to differentiate AD patients from the CN participants (SE = 82, SP = 97, AUC = 0.902, p < 0.0001), better than GMV in hippocampus (SE = 79, SP =83, AUC = 0.880, p < 0.0001) (Table 4, Supplementary Table 6). Finally, the results of the values added to the ROC curve analysis showed an improvement in the classification between the CN and AD groups using the GMV of the hippocampus and the HFC value of the insula (SE = 87, SP = 87, AUC = 0.928, p < 0.0001) (Table 4). The differentiation between MCI and AD was best using the HFC values of the insula (SE = 83, SP = 89, AUC = 0.889, p < 0.0001) (Supplementary Table 6). However, the differentiation between the CN and MCI groups was much less than between the CN and AD groups. Therefore, we need to further develop the HFC imaging sequence to use it for detecting MCI. Currently, we are developing a technique to map low-frequency conductivity in AD patients (23, 25). The diagnostic value of low-frequency conductivity is unknown.

The apparent electrical conductivity in the biological tissues is proportional to the summation of the multiplication of the ion concentrations and the electrical mobility of the charge carriers. From the HFC, which provides mixed information from the intracellular and extracellular compartments, decomposition techniques by separating the signals into the intracellular and extracellular compartments have the potential to recover the electrical properties in each separated compartment. In AD patients, both the concentration and diffusivity are increased by the demyelination and neurodegeneration of neurons replaced by CSF. Therefore, we expected increased conductivity in the AD patients compared to the CN participants. Furthermore, the previous studies showed increased metallic ions in AD patients caused by misfolding and aggregation of amyloid-beta proteins, which influenced homeostasis (5, 6) and imbalanced the Na+ and K+ ions (7, 8). These effects can be increased in HFC signals in AD patients.

As we can see in Figure 3, HFC was strongly negatively correlated with the age adjusted MMSE scores, indicating that the HFC values increased with increasing disease severity. The neuronal loss in AD patients can cause an increase in ion concentrations and diffusivity in the regions, increasing the HFC. The previous studies suggested that Aβ could form aberrant ion channels in neuronal membranes (26) and increase ion concentrations (7). However, in this study, we did not verify the relationship between increasing HFC in AD patients and the accumulation of Aβ or tau proteins. Therefore, further studies are required to investigate the relationship. We found that the HFC values were significantly correlated with age and were increased with increasing age in some areas in the brain (Table 3). Since the GMV and WMV were decreased with increasing age, the neuronal loss with aging can cause an increase in HFC in the region.