In our previous study, we evaluated the features that could differentiate GIST from leiomyoma on EUS; heterogeneity, hyperechogenic spots, a marginal halo, and higher echogenicity in comparison with the surrounding muscle layer were helpful for predicting GIST . However, judgment of these findings on EUS images is subjective; this can result in poor interobserver agreement [11, 12]. To overcome this limitation, we attempted to derive more objective findings from EUS images.
A EUS image is composed of pixels, and its echo density is expressed in brightness values from 0 (black) to 255 (white). Analysis of the brightness is, in principle, a method to evaluate the level of echogenicity (expressed as Tmean) and the degree of homogeneity (expressed as TSD). In addition, EUS images can display different characteristics in accordance with various contrasts used during an examination. Therefore, to minimize these differences, we selected the brightness of the anechoic center and outer hyperechoic rim of the EUS scope, which have the least variability, and also standardized the EUS images.
After post-standardized image analysis, both the Tmean and TSD were significantly higher in GIST than in leiomyoma and schwannoma. These results are consistent with those of previous studies that have reported higher echogenicity in comparison with the surrounding muscle layer, and heterogeneity is useful in diagnosing GIST [9, 10, 14]. In other words, we believe it is suitable to express some EUS findings as objective values after digital image analysis.
According to an ROC curve, the values of Tmean and TSD showing the best sensitivity and specificity for GIST were 65 and 75, respectively. If either Tmean ≥ 65 or TSD ≥ 75 was present, the sensitivity and specificity for predicting GIST were 94% and 80%, respectively, consistent with our previous results .
Next, we attempted to differentiate between benign and malignant GISTs on the basis of image analysis after dividing the GISTs into 2 groups (benign or malignant) according to histologic risk classification. However, we found no difference in the Tmean or TSD between benign and malignant GISTs. Previous studies have suggested that large size, exogastric growth, ulceration, cystic changes, hyperechogenic foci, and irregularity of the margin favor a diagnosis of malignant gastrointestinal mesenchymal tumor [7, 8, 15, 16]. In our previous report, only size was an independent predictor on multivariate logistic regression analysis . Therefore, there is still a limitation in predicting the malignant potential of GIST with the use of image analysis.
This study has several limitations. First, this was a retrospective study that compared EUS features between GISTs and benign mesenchymal tumors using digital image analysis. Therefore, there might have been a potential bias when retrospectively reviewing the EUS images. During the EUS examination, we obtained at least 10 endosonographic images to determine the characteristics of gastric mesenchymal tumors; we hoped this would compensate, to some degree, for the limitation of this being a retrospective study. Second, although EUS examinations were performed, patients were selected for surgery according to the clinical opinions and decisions of the medical doctors. Third, the number of patients with leiomyoma or schwannoma included in this study was small, relative to the number of those with GIST. This limitation might be due to the fact that the most common mesenchymal tumor of the stomach is GIST and that other tumors, such as leiomyoma or schwannoma, are rarely encountered in clinics. Lastly, even though we analyzed only the EUS images obtained at 7.5 MHz in order to reduce differences between the images that could be due to different frequencies, the real settings of EUS, such as gain and contrast, were different in each case, which is a limitation inherent to a retrospective study. We did attempt to standardize the EUS images on the basis of the brightness values of the anechoic center and outer hyperechoic rim of the scope. However, this attempt to standardize the EUS images will not completely overcome the limitations of a retrospective study. Therefore, prospective studies will be needed that use the same conditions of settings such as frequency, gain, and contrast.
Gastric mesenchymal tumor is often asymptomatic, and is usually discovered incidentally during upper gastrointestinal endoscopy for an unrelated condition. The main problem in asymptomatic patients is to determine whether the tumor has a malignant potential. Because GISTs have malignant potential, gastric mesenchymal tumors should not be ignored, even if they are small, if the EUS features are suggestive of GIST. Therefore, if the digital image analysis suggests a high possibility of a GIST, it would be better to attempts to obtain tissue (such as by EUS-guided fine-needle aspiration or biopsy) or to resect the tumor (such as by endoscopic or surgical resection). Further large prospective studies are required to validate our results of EUS image analysis of gastric mesenchymal tumors.