Quantitative analysis of pelvic bone marrow fat using an MRI-based machine learning method for distinguishing aplastic anaemia from myelodysplastic syndromes. Issue 6 (June 2023)

• Record Type: Journal Article
• Authors: Xiang, P.; Wu, X.; Zeng, Z.; Lin, Z.; Guo, Y.; Ma, X.; Lin, J.; Wang, W.
• Is Part Of: Clinical radiology. Volume 78:Issue 6(2023)
• Page Start: e463
• Page End: e468
• Publication Date: 2023-06
• Subjects: Medical radiology -- Periodicals; Radiotherapy -- Periodicals; Radiotherapy -- Periodicals; Radiology -- Periodicals; Societies, Medical -- Periodicals; Medical radiology; Radiotherapy; Electronic journals; Periodicals; 616.0757
• DOI: 10.1016/j.crad.2023.02.012
• Languages: English
• ISSNs: 0009-9260

Abstract

Abstract : AIM: To determine the prospect of using machine learning with magnetic resonance imaging (MRI) to identify aplastic anaemia (AA) and myelodysplastic syndromes (MDS). MATERIALS AND METHODS: This retrospective study included patients diagnosed with AA or MDS by pathological bone marrow biopsy, who underwent pelvic MRI with the iterative decomposition of water and fat with echo asymmetry and least-squares estimation quantitation (IDEAL-IQ) between December 2016 and August 2020. Based on values of right ilium fat fraction (FF) and radiomic features extracted from T1-weighted (T1W) and IDEAL-IQ images, three machine learning algorithms including linear discriminant analysis (LDA), logistic regression (LR), and support vector machine (SVM) were used to identify AA and MDS. RESULTS: A total of 77 patients were included in the study, including 37 men and 40 women, aged 20–84 years (median age 47 years). There were 21 patients with MDS (nine men and 12 women, aged 38–84 years, median age 55 years) and 56 patients with AA (28 men and 28 women, aged 20–69 years, median age 41 years). The ilium FF of patients with AA (mean ± standard deviation [SD]: 79.23 ± 15.04%) was determined to be significantly greater compared to MDS patients (mean ± SD: 42.78 ± 30.09%, p <0.001). Selecting from the machine learning models based on ilium FF, T1W imaging and IDEAL-IQ, the IDEAL-IQ-based SVM classifier model had the best predictive ability. CONCLUSION: The combination of machine learning and IDEAL-IQ technology may enable non-invasive and accurate identification of AA and MDS. Highlights: This study explored the diagnostic value of IDEAL-IQ in AA and MDS. The ilium FF of patients with AA is significantly higher than MDS. The IDEAL-IQ-based SVM classifier model has the better predictive ability than T1W.