Reproducibility of texture features with block sequential regularized expectation maximization (BSREM) reconstruction on FDG PET/MR

• Published in: The Journal of nuclear medicine (1978) Vol. 59; p. 645
• Main Authors: Nogami, Munenobu, Zeng, Feibi, Wakabayashi, Yuichi, Ueno, Yoshiko, Kanda, Tomonori, Sofue, Keitaro, Kurimoto, Takako, Kubo, Kazuhiro, Murakami, Takamichi
• Format: Journal Article
• Language: English
• Published: New York Society of Nuclear Medicine 01.05.2018
• Subjects: Attenuation; Error correction & detection; Evaluation; Feature extraction; Fluorine isotopes; Glycolysis; Histograms; Image processing; Image reconstruction; Impact analysis; Lesions; Maximization; Medical imaging; Optimization; Pathology; Patients; Point spread functions; Positron emission; Positron emission tomography; Radioisotopes; Reproducibility; Texture; Tomography; Tumors
• ISSN: 0161-5505

Purpose: A block sequential regularized expectation maximization (BSREM) reconstruction (termed Q.Clear) enables a fully convergent PET image reconstruction by controlling image noise through regularized reconstruction, and have a potential for providing more reproducible images than a conventional ordered-subsets expectation maximization (OSEM) reconstruction. Our hypothesis was that the BSREM could improve the reproducibility of texture feature on FDG PET. The purpose of the study was to assess the impact of BSREM and time-of-flight (TOF) reconstruction on texture features in patients with malignant tumors. MATERIALS AND Methods: Twenty-three patients with pathologically confirmed malignant tumors underwent F-18 FDG PET/MR (Signa PET/MR, GE healthcare) and were retrospectively evaluated. Two sets of sequential MR attenuation correction scan and PET emission scan were performed within 20 minutes, and reconstructed separately for following reproducibility studies. Each PET scan was reconstructed by OSEM with and without TOF and/ or point spread function (PSF) (OSEM, OSEM_PSF, TOFOSEM, and TOFOSEM_PSF, respectively). For BSREM reconstruction, different beta value (β=200, 400 and 600) were adopted with and without TOF (BSREM and TOFBSREM). Thirty-one texture features, four histogram indices, SUVs and total legion glycolysis (TLG) were evaluated. The reproducibility of each feature was assessed by the Bland and Altman’s analysis. RESULTS: A total of 23 lesions in 23 patients were evaluated with 10 different reconstruction methods and 35 indices including texture features. There was no significant difference in SUVs and TLG between two scans for all of the subjects (p>0.05). Six of the 35 indices showed more than 50% of the limits of agreements and was not considered as reproducible indices for all of the reconstructions. The limits of agreements between two scans for BSREM and TOFBSREM were superior to OSEM and TOFOSEM in 12 of 29 reproducible features, whereas no significant change was found in 10 features and inferior in 7 indices. Although TOFOSEM showed smaller limits of agreements than OSEM in 12 features, there was no significant relationship between reproducibility and TOF reconstruction and/or β values on BSREM reconstruction. CONCLUSIONS: BSREM is more robust in assessment of texture features than OSEM reconstruction in approximately half of the reproducible indices. TOF reconstruction seems to improve reproducibility on OSEM reconstruction; however, impact of TOF and β value in BSREM on texture features are variable.