O-190 Establishment of non-invasive prediction models for diagnosis of the subtypes and tissue composition of uterine leiomyomas by machine learning using MRI data

Abstract Study question We investigated whether machine learning models using MRI data can predict the subtypes and tissue composition of uterine leiomyomas. Summary answer Our were able to leiomyomas with high accuracy. What is known already Recently, somatic mutations in Mediator complex subunit 12 (MED12) gene found be a biomarker leiomyomas, which detected about 70% Uterine are classified into two or without MED12 mutation. These differ ratio smooth muscle cells fibroblasts amount collagen fibers. In addition, sensitivities female hormones between fibroblasts. Thus, effect therapeutic drugs (GnRH analogs selective progesterone receptor modulators) may depending on design, size, duration analyzed 90 leiomyoma nodules (MED12 mutation-positive negative = 62 28) obtained from 51 women who underwent surgery at our hospital 2020 2022. Seventy-one 49 22) assigned primary dataset establish prediction models. Nineteen 13 6) test validate model utility. Participants/materials, setting, methods For each leiomyoma, tumor signal intensity was quantified by five sequences (T2WI, ADC, T1map, T2*BOLD, MTC) for evaluating amount. After surgery, genotyping examined Trichrome staining performed quantify Using these results, we established based applying support vector classification logistic regression subtype prediction, Ridge prediction. Main results role chance The all differed significantly subtypes. cross-validation within showed that highly predictive (AUC: 0.984 0.995, respectively). validation both 1.000, both). This result higher accuracy than sequence’s cut-off value alone. On other hand, four sequence values (other ADC) significant correlation However, improve accuracy, added ADC predicted preceding predictors Support shown (R2: 0.570 0.525, respectively) 0.648 0.675, Moreover, only T2WI taken general clinical practice, as accurate sequences. Limitations, reasons caution this study, used limited one specific single center have not verified similar would under different conditions. Therefore, further training evaluation larger cohorts required use. Wider implications findings study useful such predicting drug therapy selecting drugs, GnRH modulators. Their utilities will confirmed studies involving therapy. Trial registration number applicable