| BackgroundUterine fibroids are the most common benign tumors in women,ultrasonography is currently the first-line imaging method for clinical diagnosis of fibroids because.Nevertheless,there are still more problems in diagnosing uterine fibroids with ultrasound.The first problem is the confusion between subplasmic and giant fibroids and masses of pelvic and adnexal.Second,the current lack of standardized image acquisition views and the performance differences among types of ultrasound equipment impact the accuracy of fibroid detection.In addition,the accuracy of ultrasound uterine fibroid diagnosis depends on the knowledge and experience of ultrasonographers.However,the research on deep learning-assisted medical imaging for diagnosing diseases has developed rapidly in recent years,but the current domestic and international research generally focuses on superficial organs,while relatively little research has been done in deep organs such as uterine masses.Therefore,it is necessary to develop a model for uterine fibroid detection in ultrasound images based on deep learning methods.PurposeWe established a new deep learning method to assist the ultrasound diagnosis of uterine fibroids,and then analyzed the effectiveness and feasibility of deep learning assistance methods,finally we validated its assistance to junior ultrasonographers to improve the diagnostic performance of uterine fibroids.MethodsIn this retrospective study,we collected a total of 3870 ultrasound images(2020 uterine fibroids and 1875 normal uteruses)from 667 patients with a mean age of 42.45 years± 6.23[SD]for those who received a pathologically confimed diagnosis of uterine fibroids and 570 women with a mean age of 39.24 years± 5.32[SD]without uterine lesions from Shunde Hospital between 2015 and 2020.The deep convolutional neural network was trained and developed on the training dataset(2706 images,including 1416 uterine fibroids and 1290 normal uteruses)and internal validation dataset(676 images,including 336 uterine fibroids and 340 normal uteruses).To evaluate the performance of the model on the external validation dataset(488 images,including 268 uterine fibroids and 220 normal uteruses),we assessed the diagnostic performance of the deep convolutional neural network by calculating the positive predictive value,negative predictive value,sensitivity,specificity.accuracy,and area under the receiver operating characteristic(ROC)curve for the diagnosis of the presence of uterine fibroids by junior ultrasonographers,senior ultrasonographers,deep neural convolutional neural network model,and junior ultrasonographers with the assistance of deep convolutional neural network model.Results1.Deep convolutional neural network model compared with junior ultrasonographers and senior ultrasonographersThe deep convolutional neural network model in diagnosing uterine fibroids with higher accuracy(97.27%vs.86.63%,P<0.001).sensitivity(91.79%vs.83.21%,P=0.002),specificity 97.27%vs.90.80%,P=0.005),positive predictive value(97.62%vs.91.68%,P=0.004),and negative predictive value(90.68%vs.81.61%,P=0.004)than they achieved alone.Their ability was comparable to that of senior ultrasonographers in terms of accuracy(97.27%vs.95.24%,P=0.47),sensitivity(91.79%vs.93.66%,P=0.41),specificity(97.27%vs.97.1 6%,P>0.99).positive predictive value(97.62%vs.97.57%,P=0.97),and negative predictive value(90.68%vs.92.63%,P=0.44).2.The deep convolutional neural network model aided the junior ultrasonographers compared with junior ultrasonographers and senior ultrasonographersThe deep convolutional neural network model aided the junior ultrasonographers in diagnosing uterine fibroids with higher accuracy(94.72%vs.86.63%,P<0.001),sensitivity(92.82%vs.83.21%,P=0.001),specificity(97.05%vs.90.80%,P=0.009),positive predictive value(97.45%vs.91.68%,P=0.007),and negative predictive value(91.73%vs.81.61%,P=0.001)than they achieved alone.Their ability was comparable to that of senior ultrasonographers in terms of accuracy(94.72%vs.95.24%,P=0.66),sensitivity(92.82%vs.93.66%,P=0.73),specificity(97.05%vs.97.16%,P=0.79),positive predictive value(97.45%vs.97.57%,P=0.77),and negative predictive value(91.73%vs.92.63%,P=0.75).ConclusionThe deep convolutional neural network model-assisted strategy can improve the performance of junior ultrasonographers in diagnosing uterine fibroids to a greater extent,and the deep convolutional neural network model assists junior ultrasonographers in diagnosing uterine fibroids to make them more comparable to senior ultrasonographers. |
PDF Full Text Request