Xiaoqiu Dong

Exploratory study on the enhancement of O-RADS application effectiveness for novice ultrasonographers via deep learning

The study aimed to create a deep convolutional neural network (DCNN) model based on ConvNeXt-Tiny to identify classic benign lesions (CBL) from other lesions (OL) within the Ovarian-Adnexal Reporting and Data System (O-RADS), enhancing the system's utility for novice ultrasonographers. Two sets of sonographic images of pathologically confirmed adnexal lesions were retrospectively collected [development dataset (DD) and independent test dataset (ITD)]. The ConvNeXt-Tiny model, optimized through transfer learning, was trained on the DD using the original images directly and after automatic lesion segmentation by a U-Net model. Models derived from both training paradigms were validated on the ITD for sensitivity, specificity, accuracy, and area under the curve (AUC). Two novice ultrasonographers were assessed in O-RADS with and without assistance from the model for Application Effectiveness. The ConvNeXt-Tiny model trained on original images scored AUCs of 0.978 for DD and 0.955 for ITD, while the U-Net segmented image model achieved 0.967 for DD and 0.923 for ITD; neither showed significant differences. When assessing the malignancy of lesions using O-RADS 4 and 5, the diagnostic performances of two novice ultrasonographers and senior ultrasonographer, as well as model-assisted classifications, showed no significant differences, except for one novice's low accuracy. This approach reduced classification time by 62 and 64 min. The kappa values with senior doctors' classifications rose from 0.776 and 0.761 to 0.914 and 0.903, respectively. The ConvNeXt-Tiny model demonstrated excellent and stable performance in distinguishing CBL from OL within O-RADS. The diagnostic performance of novice ultrasonographers using O-RADS is essentially equivalent to that of senior ultrasonographer, and the assistance of the model can enhance their classification efficiency and consistency with the results of senior ultrasonographer.

Prediction of benign and malignant ovarian tumors using Resnet34 on ultrasound images

AbstractObjectiveTo develop deep learning (DL) prediction models using transvaginal ultrasound (TVS), transabdominal ultrasound (TAS), and color Doppler flow imaging (CDFI) of TVS (CDFI_TVS) to automatically predict benign or malignant ovarian tumors.MethodsThis retrospective study included women with ovarian tumors who underwent ultrasound between August 2018 and October 2022. Histopathological analysis was used as a reference standard. The dataset was preprocessed by clipping, flipping, and rotating images to generate a larger, more complicated, and diverse dataset to improve accuracy and generalizability. The dataset was then divided into training (80%) and test (20%) sets. The weights of the models, modified from the residual network (ResNet) with the TVS, TAS, and CDFI_TVS images (hereafter, referred to as DLTVS, DLTAS, and DLCDFI_TVS, respectively) were developed. The area under the receiver operating characteristic curve (AUC) analysis in the test set was used to compare the predictive value of DL for malignancy.ResultsA total of 2340 images from 1350 women with adnexal masses were included. DLTVS had an AUC of 0.95 (95% CI: 0.93–0.97) for classifying malignant and benign ovarian tumors, comparable with that of DLTAS (AUC, 0.95; 95% CI: 0.91–0.98; p = 0.96) and DLCDFI_TVS (AUC, 0.88; 95% CI: 0.84–0.93; p = 0.02). Decision curve analysis indicated that DLTVS performed better than DLTAS and DLCDFI_TVS.ConclusionWe developed DL models based on TVS, TAS, and CDFI_TVS on ultrasound images to predict benign and malignant ovarian tumors with high diagnostic performance. The DLTVS model had the best prediction compared with the DLTAS and DLCDFI_TVS models.