OBJECTIVE: To evaluate the impact of a four-month training program on radiology residents\' diagnostic accuracy in assessing deep myometrial invasion (DMI) in endometrial cancer (EC) using MRI.
METHODS: Three radiology residents with limited EC MRI experience participated in the training program, which included conventional didactic sessions, case-centric workshops, and interactive classes. Utilizing a training dataset of 120 EC MRI scans, trainees independently assessed subsets of cases over five reading sessions. Each subset consisted of 30 scans, the first and the last with the same cases, for a total of 150 reads. Diagnostic accuracy metrics, assessment time (rounded to the nearest minute), and confidence levels (using a 5-point Likert scale) were recorded. The learning curve was obtained plotting the diagnostic accuracy of the three trainees and the average over the subsets. Anatomopathological results served as the reference standard for DMI presence.
RESULTS: The three trainees exhibited heterogeneous starting point, with a learning curve and a trend to more homogeneous performance with training. The diagnostic accuracy of the average trainee raised from 64 % (56 %-76 %) to 88 % (80 %-94 %) across the five subsets (p < 0.001). Reductions in assessment time (5.92 to 4.63 min, p < 0.018) and enhanced confidence levels (3.58 to 3.97, p = 0.12) were observed. Improvements in sensitivity, specificity, positive predictive value, and negative predictive value were noted, particularly for specificity which raised from 56 % (41 %-68 %) in the first to 86 % (74 %-94 %) in the fifth subset (p = 0.16). Although not reaching statistical significance, these advancements aligned the trainees with literature performance benchmarks.
CONCLUSIONS: The structured training program significantly enhanced radiology residents\' diagnostic accuracy in assessing DMI for EC on MRI, emphasizing the effectiveness of active case-based training in refining oncologic imaging skills within radiology residency curricula.

Deep myometrial invasion (≥50%) is a prognostic factor for lymph node metastases and decreased survival in endometrial cancer. There is no consensus regarding which pre/intraoperative diagnostic method should be preferred. Our aim was to explore the pattern of diagnostic methods for myometrial invasion assessment in Sweden and to evaluate differences among magnetic resonance imaging (MRI), transvaginal sonography, frozen section, and gross examination in clinical practice.
This is a nationwide historical cohort study; women with endometrial cancer with data on assessment of myometrial invasion and FIGO stage I-III registered in the Swedish Quality Registry for Gynecologic Cancer (SQRGC) between 2017 and 2019 were eligible. Data on age, histology, FIGO stage, method, and results of myometrial invasion assessment, pathology results, and hospital level were collected from the SQRGC. The final assessment by the pathologist was considered the reference standard.
In the study population of 1401 women, 32% (n = 448) had myometrial invasion of 50% of more. The methods reported for myometrial invasion assessment were transvaginal sonography in 59%, MRI in 28%, gross examination in 8% and frozen section in 5% of cases. Only minor differences were found for age and FIGO stage when comparing methods applied for myometrial invasion assessment. The sensitivity, specificity, and accuracy to find myometrial invasion of 50% or more with transvaginal sonography were 65.6%, 80.3%, and 75.8%, for MRI they were 76.9%, 71.9%, and 73.8%, for gross examination they were 71.9%, 93.6%, and 87.3%, and for frozen section they were 90.0%, 92.7%, and 92.0%, respectively.
In Sweden, the assessment of deep myometrial invasion is most often performed with transvaginal sonography, but the sensitivity is lower than for the other diagnostic methods. In clinical practice, the accuracy is moderate for transvaginal sonography and MRI.

OBJECTIVE: The identification of deep myometrial invasion (DMI) represents a fundamental aspect in patients with endometrial cancer (EC) for accurate disease staging. It can be detected on MRI using T2-weighted (T2-w), diffusion weighted (DWI) and dynamic contrast enhanced sequences (DCE). Aim of the study was to perform a multi-reader evaluation of such sequences to identify the most accurate and its reliability for the best protocol.
METHODS: In this multicenter retrospective study, MRI were independently evaluated by 4 radiologists (2 senior and 2 novice) with a sequence-based approach to identify DMI. The performance of the entire protocol was also evaluated. A comparison between the different sequences assessed by the same reader was performed using receiver operating curve and post-hoc analysis. Intraclass Correlation Coefficient (ICC) was used to assess inter- and intra-observer variability.
RESULTS: A total of 92 patients were included. The performance of the readers did not show significant differences among DWI, DCE and the entire protocol. For only one senior radiologist, who reached the highest diagnostic accuracy with the entire protocol (82,6 %), both DWI (p = 0,0197) and entire protocol (p = 0,0039) were found significantly superior to T2-w. The highest inter-observer agreement was obtained with the entire protocol by expert readers (ICC = 0,77).
CONCLUSIONS: For the detection of DMI, the performances of DWI and DCE alone and that of a complete protocol do not significantly differ, even though the latter ensures the highest reliability particularly for expert readers. In cases in which T2-w and DWI are consistent, an unenhanced protocol could be proposed.

To evaluate an MRI radiomics-powered machine learning (ML) model\'s performance for the identification of deep myometrial invasion (DMI) in endometrial cancer (EC) patients and explore its clinical applicability.
Preoperative MRI scans of EC patients were retrospectively selected. Three radiologists performed whole-lesion segmentation on T2-weighted images for feature extraction. Feature robustness was tested before randomly splitting the population in training and test sets (80/20% proportion). A multistep feature selection was applied to the first, excluding noninformative, low variance features and redundant, highly-intercorrelated ones. A Random Forest wrapper was used to identify the most informative among the remaining. An ensemble of J48 decision trees was tuned and finalized in the training set using 10-fold cross-validation, and then assessed on the test set. A radiologist evaluated all MRI scans without and with the aid of ML to detect the presence of DMI. McNemars\'s test was employed to compare the two readings.
Of the 54 patients included, 17 had DMI. In all, 1132 features were extracted. After feature selection, the Random Forest wrapper identified the three most informative which were used for ML training. The classifier reached an accuracy of 86% and 91% and areas under the Receiver Operating Characteristic curve of 0.92 and 0.94 in the cross-validation and final testing, respectively. The radiologist performance increased from 82% to 100% when using ML (p = 0.48).
We proved the feasibility of a radiomics-powered ML model for DMI detection on MR T2-w images that might help radiologists to increase their performance.