Real-time computer-aided diagnosis of focal pancreatic masses from endoscopic ultrasound imaging based on a hybrid convolutional and long short-term memory neural network model

Abstract

Differential diagnosis of focal pancreatic masses is based on endoscopic ultrasound (EUS) guided fine needle aspiration biopsy (EUS-FNA/FNB). Several imaging techniques (i.e. gray-scale, color Doppler, contrast-enhancement and elastography) are used for differential diagnosis. However, diagnosis remains highly operator dependent. To address this problem, machine learning algorithms (MLA) can generate an automatic computer-aided diagnosis (CAD) by analyzing a large number of clinical images in real-time. We aimed to develop a MLA to characterize focal pancreatic masses during the EUS procedure. The study included 65 patients with focal pancreatic masses, with 20 EUS images selected from each patient (grayscale, color Doppler, arterial and venous phase contrast-enhancement and elastography). Images were classified based on cytopathology exam as: chronic pseudotumoral pancreatitis (CPP), neuroendocrine tumor (PNET) and ductal adenocarcinoma (PDAC). The MLA is based on a deep learning method which combines convolutional (CNN) and long short-term memory (LSTM) neural networks. 2688 images were used for training and 672 images for testing the deep learning models. The CNN was developed to identify the discriminative features of images, while a LSTM neural network was used to extract the dependencies between images. The model predicted the clinical diagnosis with an area under curve index of 0.98 and an overall accuracy of 98.26%. The negative (NPV) and positive (PPV) predictive values and the corresponding 95% confidential intervals (CI) are 96.7%, [94.5, 98.9] and 98.1%, [96.81, 99.4] for PDAC, 96.5%, [94.1, 98.8], and 99.7%, [99.3, 100] for CPP, and 98.9%, [97.5, 100] and 98.3%, [97.1, 99.4] for PNET. Following further validation on a independent test cohort, this method could become an efficient CAD tool to differentiate focal pancreatic masses in real-time.

Diagnostic Value of Artificial Intelligence-Assisted Endoscopic Ultrasound for Pancreatic Cancer: A Systematic Review and Meta-Analysis

“…These variations mainly include differences in the study population, study design, interventions, and interpretations of results. In our case, the heterogeneity between studies was not significant, regarding the study design, they were eight retrospectives [ 17 , 18 , 19 , 20 , 21 , 24 , 25 , 26 ] and two prospective studies [ 22 , 23 ]. A heterogeneous element was the target used for image analysis—some of the studies used B-mode image analysis [ 18 , 20 , 23 ], from texture features [ 17 , 25 , 26 ], digital features [ 21 ], or Grey-scale pixels [ 18 ] analysis, while others made their image recognition with time-intensity curves parameters from contrast-enhanced EUS [ 20 ] or were multiparametric (B-mode, contrast, elastography) [ 25 ].…”

“…Four more articles were excluded after detailed assessment. In total, 10 studies, including 1871 patients, met our inclusion criteria [ 17 , 18 , 19 , 20 , 21 , 22 , 23 , 24 , 25 , 26 ]. The main characteristics of the studies are presented in Table 1 .…”

“…Das 2008 [ 17 ] was considered at high risk of bias for the index test because the index test results were not interpreted without knowledge of the results of the reference standard. Marya 2020 [ 20 ], Ozkan 2015 [ 21 ], Tonozuka 2021 [ 23 ], Udristoiu 2021 [ 24 ], and Zhang 2010 [ 25 ] were unclear about whether a threshold of the index test was used. Ozkan 2015 [ 21 ] was unclear about whether the reference standard results were interpreted without knowledge of the results of the index test and about whether there was an appropriate interval between index and reference test.…”

“…On the other hand, only three studies were included in the subgroup of deep learning-based CAD [ 18 , 23 , 24 ]. As we expected, deep learning had a higher accuracy of diagnosis than conventional type in PC patients, and an impressive diagnostic efficacy was found with the AUC of 0.97.…”

“…There were four types of AI algorithms used to generate the automatic, real-time diagnosis of PC: artificial neural network (ANN), with three studies [ 17 , 21 , 22 ]; convolutional neural network (CNN), with four studies [ 18 , 20 , 23 , 24 ]; and support vector machine (SVM), with two studies [ 25 , 26 ]; and there was only one basic machine learning model, unclassified [ 19 ]. Different as they are, algorithms have advantages of their own.…”