Assessment of AI-Enhanced Quantitative Volumetric MRI with Semi-Quantitative Analysis in 18F-FDG Metabolic Imaging for Alzheimer's Diagnosis.

Introduction: The integration of Magnetic Resonance Imaging (MRI) and Positron Emission Tomography (PET) data, coupled with advancements in artificial intelligence (AI), has significantly enhanced the semi-quantitative aspects of morphologic imaging. This study systematically scrutinizes the diagnostic precision of quantitative MRI, employing the AI tool "Heuron AD" for automated volumetric measurements. The primary objective is to discern the clinical efficacy of Heuron AD and its potential for early Alzheimer's disease (AD) detection in 18F-FDG PET metabolic imaging.

15 AD patients (6 men, 9 women; mean age ± SD: 70.4 ± 9.72) and 15 normal subjects (5 men, 10 women; mean age ± SD: 66.73 ± 7.99) were enrolled, aged 54-84 years and 51-80 years, respectively. PET images were acquired at 50-60 minutes after 18F-FDG injection. The Heuron AD software was used to extract brain volume. P-mod software was used to compute standardized uptake value ratio (SUVr) and software was used to calculate Z scores to compare FDG uptake. Stata software was employed to find the correlation between extracted brain volume, SUVr, and Z scores. Two experienced radiologists, blinded to clinical information, performed visual interpretations of the MRI images. AD patients were identified by using the ATN system. MRI images were obtained using Siemens mMR PET/MRI.

Methods: 15 AD patients (6 men, 9 women; mean age ± SD: 70.4 ± 9.72) and 15 normal subjects (5 men, 10 women; mean age ± SD: 66.73 ± 7.99) were enrolled, aged 54-84 years and 51-80 years, respectively. PET images were acquired at 50-60 minutes after 18F-FDG injection. The Heuron AD software was used to extract brain volume. P-mod software was used to compute standardized uptake value ratio (SUVr) and software was used to calculate Z scores to compare FDG uptake. Stata software was employed to find the correlation between extracted brain volume, SUVr, and Z scores. Two experienced radiologists, blinded to clinical information, performed visual interpretations of the MRI images. AD patients were identified by using the ATN system. MRI images were obtained using Siemens mMR PET/MRI.

Results: The software yielded a balanced sensitivity and specificity for detection of Alzheimer's-related changes in hippocampal volume with a good accuracy (73.33%) compared to 66.67% accuracy (46.67% sensitivity, 86.67% specificity) by radiologists. Brain volume examination in the identified AD signature area also revealed 66.67% accuracy (33.33% sensitivity, 100% specificity). The atrophy index, a measure of shrinkage, of hippocampal volume using Heuron AD, notably correlated with FDG SUVr in the hippocampus. The hippocampal Z-score moderately correlated with the atrophy index. There was also moderate correlation of atrophy index from Heuron AD’s brain signature area with FDG SUVr. However, no correlation was observed between the atrophy index from the brain signature area and the Z-score The diagnostic performance, indicated by the area under the curve (AUC), was 0.83 for hippocampal atrophy, 0.81 for the AD signature area, 0.82 for SUVr, and 0.59 for Z-score calculation.

Conclusions: The software exhibits promising potential for detection of Alzheimer's-related hippocampal volume loss, outperforming radiologists’ interpretations with an accuracy of 73.33% and the highest AUC compared with all methods. Notably, there is a strong correlation of Heuron AD's hippocampal atrophy index and FDG SUVr values, highlighting its sensitivity to subtle atrophy. The use of AI with radiologists collaboratively holds the promise of enhancing precision and sensitivity in the diagnosis of Alzheimer’s Disease. However, ongoing development remains paramount for optimizing the software, particularly in refining the analysis method of signature brain AD area where enhancements in sensitivity may be imperative.