AUTHOR=Li Yali , Jiang Yaojun , Yu Xi , Ren Binbin , Wang Chunyu , Chen Sihui , Ma Duoshan , Su Danyang , Liu Huilong , Ren Xiangyang , Yang Xiaopeng , Gao Jianbo , Wu Yan TITLE=Deep-learning image reconstruction for image quality evaluation and accurate bone mineral density measurement on quantitative CT: A phantom-patient study JOURNAL=Frontiers in Endocrinology VOLUME=Volume 13 - 2022 YEAR=2022 URL=https://www.frontiersin.org/journals/endocrinology/articles/10.3389/fendo.2022.884306 DOI=10.3389/fendo.2022.884306 ISSN=1664-2392 ABSTRACT=Background and Purpose: To investigate image quality and accurate bone mineral density (BMD) on quantitative CT (QCT) for osteoporosis screening by deep-learning image reconstruction (DLIR) based on a multi-phantom/patient study. Materials and Methods: Modulation function test (MTF), noise power spectrum (NPS) and image noise were evaluated for physical image quality on Caphan 500. HA inserts was measured on European Spine Phantom (ESP). Further image analysis and BMD quantification were performed on patients for clinical evaluation. Results: For 3 levels of DLIR, 50% MTF was about 4.50 (lp/cm), better than FBP (4.12 lp/cm) and ASiR-V50% (4.00 lp/cm). The 2 or 3mm low-contrast object was clearly resolved at a 0.5% contrast level or at FBP and ASiR-V50%. It showed an obviously lower NPS peak frequency and noise level, shifting towards lower spatial frequency. L1, L2 and L3 were slightly lower than that of nominal HA inserts (44.80±1.27, 95.92±1.33, 194.88±1.85 versus 50.20, 100.60, 199.20mg/cm3) with an relative accuracy error of 11.61±3.25%, 4.08±1.33% and 2.60%±0.78%. The prevalence of osteoporosis and vertebral fracture in women and men were 74 (24.67%) versus 49 (11.79%) and 26 (8.67%) versus 22 (5.29%), respectively. DLIR-M and DLIR-H were scored significantly better than ASiR-V50% in image noise (4.83±0.34, 4.50±0.50 versus 4.17±0.37), image contrast (4.67±0.73, 4.50±0.70 versus 3.80±0.99), small structure visibility (4.83±0.70, 4.17±0.73 versus 3.83±1.05), image sharpness (3.83±1.12, 3.53±0.90 versus 3.27±1.16) and artifacts (3.83±0.90, 3.42±0.37 versus 3.10±0.83). The overall image quality, CT value, image noise, SNR, CNR, and image artifacts were outperformed for DLIR at high strength levels compared with ASiR-V50% and FBP (p<0.001), whilst it was not significant different between DLIR-L and ASiR-V50% (p>0.05). Conclusion: Image quality with DLIR was high-qualified and BMD was accurate, indicating a great clinical utility in osteoporosis screening.