Machine learning for predicting full-count FDG PET brain images from low-count acquisitions in suspected dementia: a clinical and quantitative evaluation

Objective. Artificial intelligence methods for denoising low-count (LC) FDG positron emission tomography (PET) brain images are usually evaluated using image quality metrics alone, with limited direct clinical assessment, particularly in suspected dementia. This study evaluated a machine-learning image quality transfer (IQT) method for predicting full-count FDG PET brain images from LC acquisitions using both quantitative metrics and blinded clinical assessment. Approach. Forty-one FDG PET/CT patients with suspected dementia were retrospectively included, with LC images simulated using 5% of list-mode data. An IQT random forest model employing patch-wise regression was evaluated using image quality metrics, regional (Formula presented) (Formula presented) -scores, and blinded radiologist assessment against standard-count references. Main results. AI-predicted images showed an average peak signal-to-noise ratio improvement of approximately 4 dB and an approximate 36% reduction in root mean square error compared with LC images, while also outperforming a classical non-local means denoising filter. Clinically, uninterpretable scans were reduced to 0% for each reader (0/10 cases), down from 20% (2/10 cases) and 50% (5/10 cases) respectively, with a shift from tentative to confident agreement with the reference standard reports. Significance. In patients with suspected dementia, where motion and limited tolerance of long acquisitions are common, this study demonstrates the potential of an IQT method to improve the clinical usability of LC FDG PET scans.