Digital Image Analysis (DIA) Is a Useful Adjunct to Endosonographic Diagnosis of Submucosal Lesions (SML) of the Gastro-Intestinal Tract

Abstract

Establishing definitive diagnosis of gastrointestinal SML remains a challenge. Information from EUS further delineates the lesions, but histologic data remain the gold standard. To enhance the diagnostic capability of EUS, with the hope to circumvent the need for tissue acquisition, EUS images were digitally analyzed using the technique described pervious by us to discern any parameters that can differentiate clinically significant lesions such as, gastrointestinal stromal cell tumors (GIST) and carcinoids from more benign lesions such as such as, lipoma. Objective: To evaluate the role of DIA of EUS images of SML. Methods: Patients with histopathologically confirmed SML who had undergone EUS evaluation were identified from our database. Representative regions of interest (ROIs) were digitally selected from their EUS images [Gastrointest Endosc 2006;63(5): AB 256, Gastrointest Endosc 2007;65(5):AB 103, & Das, Gastrointest Endosc 2008 (in press)]. Texture analyses were performed using histogram, run-length and co-occurrence matrix (measuring runs of pixels and distribution of pairs of pixels), gradient analysis (spatial distribution of pixels), auto-regressive analysis (measurement of local interaction amongst pixels of different grey level values) and also wavelet analysis (which measures parameters of spatial frequency). Results: A total of 106, 111 and 124 ROIs were selected from EUS images of 8, 10 and 28 patients with lipoma, carcinoids and GIST, respectively. Principal component analysis (PCA) was used for data reduction and later a neural network (ANN) based predictive model was built, trained and validated using the extracted texture features for classification. The performance characteristics of the ANN model in classifying lipoma, carcinoids and GIST is shown in Table. Three types of features of pixel distribution had high discriminatory power: Kurtosis (measure of the steepness of the pixel distribution), angular second momentum and horizontal and vertical grey level non-uniformity (parameters estimating the second order joint conditional probability density functions of spatial distribution of two neighboring pixels). Conclusion: DIA of texture features of EUS images is a useful non-invasive adjunct to endosonographic diagnosis of submucosal lesions. Tabled 1 SML Sensitivity Specificity PPV NPV AUC ∗ AUC of ROC curve. Lipoma 0.90 0.77 0.64 0.95 0.92 Carcinoid 0.79 0.81 0.67 0.89 0.86 GIST 0.81 0.79 0.74 0.86 0.89 ∗ AUC of ROC curve. Open table in a new tab