Abstract
Objectives: The aim of this study was to construct a diagnostic model from MRI features to distinguish complex leiomyomas/degenerating fibroids (DF) from leiomyosarcoma (LMS). Methods: A retrospective case-controlled study was performed comparing MRI features of patients with pathologically proven DF or LMS. MRI in 42 patients with DF (control group) and 46 with LMS (study group) was used to generate a diagnostic model. Imaging features reported in the literature to distinguish these two entities were scored for each uterine mass by two radiologists unaware of the histological diagnosis. Inter observer variation and univariate analysis was undertaken. Imaging characteristics identified on univariate analysis were used to build a multi-variable diagnostic model and sensitivity and specificity of this model calculated. Results: Taking the features identified on the univariate analysis, the final diagnostic model was based on AP length (p = .053), intermediate T2 signal (IT2), volume (p = .002), and nodular border (p = .001). When the model was implemented back into the training dataset it demonstrated a sensitivity of 70.7%, and a specificity of 76.2%. The sensitivity and specificity of radiologist suspicion score was 74.7% and 70.4%. In addition, morphological features showed only poor or moderate inter observer agreement at best. Conclusions: Morphological MRI imaging features alone are not sufficient to obviate the need for pathological confirmation prior to non-surgical management of complex uterine mass lesions. Trial registration: IRAS project ID 251778 Protocol number: CCR 4992 REC reference 19/YH/0134 Date of HRA approval: 29.4.19.