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, 64 Cu-DOTATATE PET, is part of the diagnostic work-up of patients with neuroendocrine neoplasms (NENs), resulting in high accumulated radiation doses. Scanrelated 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 64 Cu-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 64 Cu-DOTATATE PET could be restored using artificial intelligence (AI). Methods: We implemented a parameter-transferred Wasserstein generative adversarial network for patients with NENs on simulated low-dose 64 Cu-DOTA-TATE 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 (n 5 33) to allow comparison of all foci on a 1:1 basis. Detected foci were scored (C 1 , definite lesion; C 0 , lesion-suspicious focus) and matched with PET 100% as the reference. True-positive (TP), false-positive (FP), and false-negative (FN) lesions were assessed. Results: 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/(TP1FN)) and a false discovery rate (FP/(TP1FP)) of 66% (78/118) and 32% (37/115), respectively. In 62% (23/37) of cases, the FP lesion was scored C 1 , suggesting a definite lesion. Conclusion: PET AI improved visual similarity with PET 100% compared with PET 25% , and PET AI and PET 100% had similar Likert scaledefined 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.