The increasing prevalence of neurodegenerative diseases, such as Alzheimer's dementia and Parkinson's disease, has necessitated the development of non-invasive diagnostic and treatment assessment tools, with brain imaging playing a pivotal role in this endeavor. PET has emerged as a valuable imaging technique for visualizing the functional and molecular characteristics of the brain, offering essential insights into the pathophysiology of neurodegenerative diseases. In particular, the ability to visualize amyloid beta protein deposition in Alzheimer’s dementia has significantly contributed to its diagnosis and treatment planning, altering the diagnosis in a substantial proportion of cases based on the results of amyloid PET.

However, the clinical utility of brain PET imaging, including amyloid PET, is hampered by the subjectivity and variability inherent in visual interpretation [17]. This study addresses these limitations by exploring quantitative parameters derived from PET images, which can provide objective information about disease progression and enhance patient management. While several methods for obtaining quantitative parameters from brain PET images exist, they face challenges in standardization, reliance on additional imaging modalities such as MRI, and the complexity of the quantification process.

Spatial normalization of brain PET images, a crucial step in quantification, facilitates the comparison of images across individuals and studies. Spatial normalization also plays a pivotal role in enabling quantitative analysis, as it transforms individual PET images into a common stereotactic space, allowing for objective regional analysis. This study focuses on evaluating the clinical performance of BTXBrain-Amyloid, an AI-based spatial normalization algorithm, and comparing it with traditional methods using SPM.

The results of this study demonstrate the superiority of BTXBrain-Amyloid in spatial normalization accuracy compared to the traditional SPM method, which utilized paired MR images for co-registration. The normalized mutual information values consistently indicated that BTXBrain achieved higher similarity to the standard template, thereby improving the accuracy and reliability of PET spatial normalization. Furthermore, the reduced standard deviation of normalized mutual information in BTXBrain suggests greater consistency and reproducibility compared to SPM, which may be particularly important in multicenter and longitudinal studies. Furthermore, the impact of spatial normalization accuracy on the subsequent voxel-level statistical analysis was evident. When using BTXBrain for spatial normalization, statistically significant differences were observed in regions such as the dorsal caudate and thalamus, which were not detected when employing SPM. This underscores the importance of accurate spatial normalization, particularly in deep brain regions, for improving the sensitivity of detecting group differences and increasing the statistical power of analyses.

The BTXBrain employs a sophisticated approach to nonlinearly register input PET images by utilizing estimated deformation fields to align them with the MNI 152 template. This template, derived from young, healthy individuals (with a mean age of 25.02 and a standard deviation of 4.9), serves as a standardized reference space. Therefore, the deformations may be substantial when input images exhibit significant atrophic regions. To mitigate the potential distortion of images, regularization techniques can be incorporated during training, or the resulting deformation fields can be smoothed. In addition, the image deformation is conducted while preserving the PET pixel intensity which corresponds to radiotracer concentration. Therefore, the quantification results derived from the normalized image using predefined atlases remains accurate. While Kang et al. have demonstrated the robustness of the BTXBrain through internal and external datasets [15], further investigations are warranted to validate its accuracy across diverse image sets and cohorts.

The clinical performance of both BTXBrain and SPM-based PET quantification methods in distinguishing between amyloid-positive and negative groups was assessed through regional SUVR values. Notably, BTXBrain demonstrated lower standard deviations of SUVR for both groups, indicating greater precision in its measurements. ROC curve analysis further highlighted the superior discriminative ability of BTXBrain, with a higher AUC compared to SPM. PR curve analysis accounted for the imbalanced group distribution in this study, further supports the superior performance of BTXBrain.

Beyond improved accuracy and discrimination, BTXBrain offers a substantial advantage in terms of efficiency. The PET quantification process using BTXBrain was remarkably faster, taking only 29 s on average, compared to more than 10 min required by the SPM method, which includes additional steps such as PET and MRI reregistration.

The remarkable performance of BTXBrain-Amyloid in this study illustrates the transformative potential of AI technology in the field of medical imaging. AI-based solutions, like BTXBrain, not only enhance the accuracy, reproducibility, and efficiency of image analysis but also have the potential to revolutionize clinical practice. In the field of medical imaging medical imaging, AI has emerged as a valuable tool, assisting healthcare professionals in tasks ranging from image interpretation and diagnosis to treatment planning and outcome prediction [18,19,20]. AI can swiftly analyze vast datasets, detect subtle patterns, and provide quantitative insights that were previously unattainable through manual methods [21,22,23,24,25,26,27]. As demonstrated in this study, AI-based spatial normalization not only simplifies the quantification process but also improves the sensitivity of detecting disease-related changes, ultimately leading to more precise and personalized patient care.

In summary, this study underscores the clinical potential of BTXBrain-Amyloid as an AI-powered quantification tool for amyloid PET imaging. It not only outperforms traditional methods in terms of accuracy, consistency, and efficiency but also enhances the sensitivity of detecting regional differences in deep brain structures critical for neurodegenerative disease research and diagnosis. The improved precision and discriminative power of BTXBrain in classifying amyloid-positive and negative cases further highlight its clinical relevance in the assessment and management of patients with neurodegenerative diseases. Ultimately, the adoption of such advanced AI-driven tools may revolutionize the field of neuroimaging, aiding in the early diagnosis and effective treatment of these devastating conditions.