An Investigation of Lesion Detection Accuracy for Artificial Intelligence–Based Denoising of Low-Dose⁶⁴Cu-DOTATATE PET Imaging in Patients with Neuroendocrine Neoplasms

Abstract

Frequent somatostatin receptor PET, for example, ⁶⁴Cu-DOTATATE PET, is part of the diagnostic work-up of patients with neuroendocrine neoplasms (NENs), resulting in high accumulated radiation doses. Scan-related radiation exposure should be minimized in accordance with the as-low-as-reasonably achievable principle, for example, by reducing injected radiotracer activity. Previous investigations found that reducing ⁶⁴Cu-DOTATATE activity to below 50 MBq results in inadequate image quality and lesion detection. We therefore investigated whether image quality and lesion detection of less than 50 MBq of ⁶⁴Cu-DOTATATE PET could be restored using artificial intelligence (AI). <b>Methods:</b> We implemented a parameter-transferred Wasserstein generative adversarial network for patients with NENs on simulated low-dose ⁶⁴Cu-DOTATATE PET images corresponding to 25% (PET_25%), or about 48 MBq, of the injected activity of the reference full dose (PET_100%), or about 191 MBq, to generate denoised PET images (PET_AI). We included 38 patients in the training sets for network optimization. We analyzed PET intensity correlation, peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and mean-square error (MSE) of PET_AI/PET_100% versus PET_25%/PET_100% Two readers assessed Likert scale-defined image quality (1, very poor; 2, poor; 3, moderate; 4, good; 5, excellent) and identified lesion-suspicious foci on PET_AI and PET_100% in a subset of the patients with no more than 20 lesions per organ (<i>n</i> = 33) to allow comparison of all foci on a 1:1 basis. Detected foci were scored (C₁, definite lesion; C₀, lesion-suspicious focus) and matched with PET_100% as the reference. True-positive (TP), false-positive (FP), and false-negative (FN) lesions were assessed. <b>Results:</b> For PET_AI/PET_100% versus PET_25%/PET_100%, PET intensity correlation had a goodness-of-fit value of 0.94 versus 0.81, PSNR was 58.1 versus 53.0, SSIM was 0.908 versus 0.899, and MSE was 2.6 versus 4.7. Likert scale-defined image quality was rated good or excellent in 33 of 33 and 32 of 33 patients on PET_100% and PET_AI, respectively_. Total number of detected lesions was 118 on PET_100% and 115 on PET_AI Only 78 PET_AI lesions were TP, 40 were FN, and 37 were FP, yielding detection sensitivity (TP/(TP+FN)) and a false discovery rate (FP/(TP+FP)) of 66% (78/118) and 32% (37/115), respectively. In 62% (23/37) of cases, the FP lesion was scored C₁, suggesting a definite lesion. <b>Conclusion:</b> PET_AI improved visual similarity with PET_100% compared with PET_25%, and PET_AI and PET_100% had similar Likert scale-defined image quality. However, lesion detection analysis performed by physicians showed high proportions of FP and FN lesions on PET_AI, highlighting the need for clinical validation of AI algorithms.