Stott W, et al. – The objective of this study is to independently evaluate self-reported visualization rates(srVR) utilizing expert review and develop software capable of monitoring it. These outcomes recommend that in a significant proportion of TVS annual screens the sonographers may have mistaken other structures for normal ovaries. It is uncertain whether or not this influenced the sensitivity and stage at detection of ovarian cancer in the ultrasound arm of UKCTOCS, but they conclude quality control(QC) metrics based on self–reported visualization of normal ovaries are unreliable. The classifier demonstrates some potential for addressing this problem, though further investigation is required.

Methods

• Images from 1,000 examinations randomly selected from 68,951 archived annual TVS exams undertaken between 2008–2011 where the ovaries were reported as ‘seen and normal’ were reviewed by a single expert.
• Software was developed to recognize exact images used to measure ovaries by measuring caliper dimensions and matching them to that recorded by the sonographer.
• A logistic regression classifier to ascertain visualization was trained and validated utilizing ovarian dimension and visualization data reported by the expert.

Results

• In this study, the expert affirmed both ovaries were visualized (cVR–Both) in 50.2%(502/1000) of the exams.
• The software distinguished the measurement image in 534 exams which were split 2:1:1 providing training, validating and testing data.
• Classifier accuracy on validation data was 70.9%(CI–95% 70.0,71.8).
• Investigation of test information (133 exams) resulted in sensitivity of 90.5%(CI–95% 80.9,95.8) and specificity of 47.5%(CI–95% 34.5,60.8) in detecting expert confirmed cVR–Both.