4070
A comparative study on MRI-based radiomics model selection for a high-risk cytogenetics prediction in multiple myeloma1peking university third hospital, beijing, China
Synopsis
We aim to develop a radiomics model based on MRI to predict high-risk cytogenetics (HRC) status in patients with multiple myeloma (MM) and identify optimal machine learning methods. We retrospectively analyzed 89 patients’ (37 HRC, 52 non-HRC) radiomics features extracted from fat suppression T2W and TIW image. The following classification methods, including, support vector machine, random forest, logistic regression (LR) and decision tree were used to construct radiomics models. LR model showed the highest performance with an AUC of 0.82 ± 0.02. Radiomics model based on LR classifier can be used to predict HRC status in patients with MM effectively.
INTRODUCTION
High-risk cytogenetic (HRC) can influence disease course, response to therapy and prognosis in multiple myeloma (MM) (1). The research for a noninvasive method to diagnose HRC has become an area of intense research interest. Radiomics method can obtain hundreds of features from the images to analysis in addition to the advantage of simple, non-invasive, safe and repeatable operation of traditional imaging. Different machine learning methods may show different performance when applied in different organs (2). Therefore, it is important to choose a suitable machine learning method for a radiomics model. The aim of this study was to develop and validate a radiomics model based on MRI in prediction of high-risk cytogenetics (HRC) status in patients with MM and identify optimal machine learning methods.METHODS
This retrospective analysis included 89 patients (37 with HRC, 52 with non-HRC). Radiomics features were extracted from two sequences of spine MRI, including fat suppression T2W image and TIW image. Three feature selection sequential steps, including variance threshold, SelectKBest, least absolute shrinkage and selection operator (LASSO), and four classification methods, including, support vector machine (SVM), random forest (RF), logistic regression (LR) and decision tree (DT) were used to distinguish HRC and non-HRC status. The prediction performances of different models were assessed by comparison of the Area Under the Curve (AUC) with DeLong’s test. Validation was performed with five-fold external cross-validation.RESULTS
In our machine learning model, four classifiers were tested and their heat map of AUC were shown in Fig. 1. In general, no matter single sequence or combination of two sequences, the classifier LR had the best performance indicated by a darkest red color, followed SVM indicated by the darker red color.For single sequence, the performance of FS-T2WI was better than T1WI (Fig. 2). Radiomics model based on combined two sequences (T1WI and T2WI) using LR classifier yielded the highest AUC of 0.82 ± 0.02 (sensitivity 84.5%; specificity 68.1%; YI 0.52; accuracy 74.7%).DISCUSSION
Previous study demonstrated that different organ may have its most suitable classifier to contrast a model, and LR may be effective in sacral tumors (2), soft-tissue masses of the extremities and pelvic (3, 4), pulmonary solid nodules (5) and breast cancer (6). The classifier LR had the best performance whether single sequence or combination sequences were used to analysis in this study. The performance of combination sequences was better than single sequence.CONCLUSION
The machine learning method of LR was superior to other classifier methods in assessing HRC status. Radiomics model based on combined sequences use LR classifier could be used to differentiate HRC and non-HRC status in patients with MM effectively.Acknowledgements
No acknowledgement found.References
(1) Rajkumar SV. Multiple myeloma: 2020 update on diagnosis, risk-stratification and management. Am J Hematol. 2020 May;95(5):548-567
(2) Wang X, Wan Q, Chen H, et al. Classification of pulmonary lesion based on multiparametric MRI: utility of radiomics and comparison of machine learning methods. Eur Radiol. 2020 Aug;30(8):4595-4605.
(3) Yin P, Mao N, Liu X, et al. Can clinical radiomics nomogram based on 3D multiparametric MRI features and clinical characteristics estimate early recurrence of pelvic chondrosarcoma? J Magn Reson Imaging. 2020 Feb;51(2):435-445.
(4) Wang H, Nie P, Wang Y, et al. Radiomics nomogram for differentiating between benign and malignant soft-tissue masses of the extremities. J Magn Reson Imaging. 2020 Jan;51(1):155-163
(5) Shen Y, Xu F, Zhu W, et al. Multiclassifier fusion based on radiomics features for the prediction of benign and malignant primary pulmonary solid nodules. Ann Transl Med. 2020 Mar;8(5):171
(6) Mao N, Yin P, Wang Q, et al. Added Value of Radiomics on Mammography for Breast Cancer Diagnosis: A Feasibility Study. J Am Coll Radiol. 2019 Apr;16(4 Pt A):485-491.
Figures
