109 adenomyosis (12,34,212). Overall, differences in baseline characteristics between the two datasets highlight the need for external validation studies to confirm the generalizability and reliability of the prediction model Furthermore, the diagnostic accuracy of radiology reports of the MRIs in our population is warrants attention. In this external validation study, the diagnostic accuracy of radiology reports for MRIs was found to be 63.6%. Similarly, the internal validation study showed an overall diagnostic accuracy of 59.4% for the radiology reports (204). These percentages of diagnostic accuracy appear to be suboptimal compared to commonly reported values in literature (97). It is possible that the radiologists did not actively assess for adenomyosis if it was explicitly requested in the MRI application. A potential solution could be to involve gynaecologists in the assessment of the MRI parameters that are required for the model, and in the assessment of pelvic MRIs in general. This could enhance the efficiency of the process, as conducting the measurements required for this model can be learned and completed in a relatively short period of time. The developed model showed good performance in the original dataset with an AUC of 0.776. After external validation, the performance of the model reached an AUC of 0.831, which indicates that the model has good to excellent discriminative ability. This means that the model is able to predict with high accuracy which patients are more likely to have histopathological adenomyosis (207). Typically, the performance of a logistic prediction model is better on the original dataset than the performance of this model on a new dataset . Patient selection for the developed dataset was between 2007 to 2022, while patient selection for the external dataset was between 2014 to 2022. This more recent patient cohort could be an explanation for the slightly better model performance in the external validation group. The general quality of the MRIs could have been better because they were more recent. Moreover, potentially the recognition of adenomyosis by the pathologist may have increased in recent years. This study also found a higher prevalence of uterine fibroids in the group without the diagnosis of adenomyosis compared to the adenomyosis group. This greater presence of uterine fibroids may have created a more homogenous group and contributed to the model's ability to better distinguish between these two groups. However, it is important to note that this is only a hypothesis.
RkJQdWJsaXNoZXIy MTk4NDMw