Background and Aims: Endoscopic ultrasonography (EUS) is a useful diagnostic modality for evaluating gastric mesenchymal tumors; however, differentiating gastrointestinal stromal tumors (GISTs) from benign mesenchymal tumors such as leiomyomas and schwannomas remains challenging. For this reason, we developed a convolutional neural network computer-aided diagnosis (CNN-CAD) system that can analyze gastric mesenchymal tumors on EUS images. Methods: A total of 905 EUS images of gastric mesenchymal tumors (pathologically confirmed GIST, leiomyoma, and schwannoma) were used as a training dataset. Validation was performed using 212 EUS images of gastric mesenchymal tumors. This test dataset was interpreted by three experienced and three junior endoscopists. Results: The sensitivity, specificity, and accuracy of the CNN-CAD system for differentiating GISTs from non-GIST tumors were 83.0%, 75.5%, and 79.2%, respectively. Its diagnostic specificity and accuracy were significantly higher than those of two experienced and one junior endoscopists. In the further sequential analysis to differentiate leiomyoma from schwannoma in non-GIST tumors, the final diagnostic accuracy of the CNN-CAD system was 75.5%, which was significantly higher than that of two experienced and one junior endoscopists. Conclusions: Our CNN-CAD system showed high accuracy in diagnosing gastric mesenchymal tumors on EUS images. It may complement the current clinical practices in the EUS diagnosis of gastric mesenchymal tumors.

Endoscopic ultrasonography (EUS) is a valuable imaging tool for evaluating subepithelial lesions in the stomach. However, there are few studies on differentiation between gastrointestinal stromal tumors (GISTs) and benign mesenchymal tumors, such as leiomyoma or schwannoma, with the use of EUS. In addition, there are limitations in the analysis of the characteristic features of such tumors due to poor interobserver agreement as a result of subjective interpretation of EUS images. Therefore, the aim of this study was to evaluate the role of digital image analysis in distinguishing the features of GISTs from those of benign mesenchymal tumors on EUS. We enrolled 65 patients with histopathologically proven gastric GIST, leiomyoma or schwannoma on surgically resected specimens who underwent EUS examination at our endoscopic unit from January 2007 to September 2010. After standardization of the EUS images, brightness values including the mean (T_mean), indicative of echogenicity, and the standard deviation (T_SD), indicative of heterogeneity, in the tumors were analyzed. The T_mean and T_SD were significantly higher in GIST than in leiomyoma and schwannoma (p < 0.001). However, there was no significant difference in the T_mean or T_SD between benign and malignant GISTs. The sensitivity and specificity were almost optimized for differentiating GIST from leiomyoma or schwannoma when the critical values of T_mean and T_SD were 65 and 75, respectively. The presence of at least 1 of these 2 findings in a given tumor resulted in a sensitivity of 94%, specificity of 80%, positive predictive value of 94%, negative predictive value of 80%, and accuracy of 90.8% for predicting GIST. Digital image analysis provides objective information on EUS images; thus, it can be useful in diagnosing gastric mesenchymal tumors.

Accurate diagnosis of acute appendicitis from abdominal ultrasound is a challenging task, since traditional sonographic diagnostic criteria for appendicitis, such as diameter, compressibility, and wall thickness, rely on complete identification or visualization of the appendix and the diagnosis is frequently operator subjective. In this paper, we propose a robust automatic segmentation method for inflamed appendix identification to mitigate abovementioned difficulties. We use outlier rejection fuzzy c-means clustering (FCM) algorithm within a double-layered learning structure to extract the target inflamed appendix area. The proposed method extracts the target appendix in 98 cases out of 100 test images, which is far better than traditional FCM, standard outlier FCM, and double-layered learning with FCM in correct extraction rate. Furthermore, we investigate the outlier rejection effect and double layered learning effect by comparing our proposed method with standard double-layered FCM and the standard outlier-rejection FCM. In this comparison, the proposed method exhibits robust segmentation results in accuracy, precision, and recall by 2.5~5.6% over two standard methods in quality with human pathologists’ marking as the ground truth.