Deep learning for pancreatic diseases based on endoscopic ultrasound: A systematic review

Bald eagle-optimized transformer networks with temporal-spatial mid-level features for pancreatic tumor classification

The classification and diagnosis of pancreatic tumors present significant challenges due to their inherent complexity and variability. Traditional methods often struggle to capture the dynamic nature of these tumors, highlighting the need for advanced techniques that improve precision and robustness. This study introduces a novel approach that combines temporal-spatial mid-level features (CTSF) with bald eagle search (BES) optimized transformer networks to enhance pancreatic tumor classification. By leveraging temporal-spatial features that encompass both spatial structure and temporal evolution, we employ the BES algorithm to optimize the vision transformer (ViT) and swin transformer (ST) models, significantly enhancing their capacity to process complex datasets. The study underscores the critical role of temporal features in pancreatic tumor classification, enabling the capture of changes over time to improve our understanding of tumor progression and treatment responses. Among the models evaluated-GRU, LSTM, and ViT-the ViTachieved superior performance, with accuracy rates of 94.44%, 89.44%, and 87.22% on the TCIA-Pancreas-CT, Decathlon Pancreas, and NIH-Pancreas-CT datasets, respectively. Spatial features extracted from ResNet50, VGG16, and ST were also essential, with the ST model attaining the highest accuracy of 95.00%, 95.56%, and 93.33% on the same datasets. The integration of temporal and spatial features within the CTSF model resulted in accuracy rates of 96.02%, 97.21%, and 95.06% for the TCIA-Pancreas-CT, Decathlon Pancreas, and NIH-Pancreas-CT datasets, respectively. Furthermore, optimization techniques, particularly hyperparameter tuning, further enhanced performance, with the BES-optimized model achieving the highest accuracy of 98.02%, 98.92%, and 98.89%. The superiority of the CTSF-BES approach was confirmed through the Friedman test and Bonferroni-Dunn test, while execution time analysis demonstrated a favourable balance between performance and efficiency.

A semi-supervised convolutional neural network for diagnosis of pancreatic ductal adenocarcinoma based on EUS-FNA cytological images

BACKGROUND

The cytological diagnostic process of EUS-FNA smears is time-consuming and manpower-intensive, and the conclusion could be subjective and controversial. Moreover, the relative lack of cytopathologists has limited the widespread implementation of Rapid on-site evaluation (ROSE) presently. Therefore, this study aimed to establish an AI system for the detection of pancreatic ductal adenocarcinoma (PDAC) based on EUS-FNA cytological images.

METHODS

We collected 3213 unified magnification images of pancreatic cell clusters from 210 pancreatic mass patients who underwent EUS-FNA in four hospitals. A semi-supervised CNN (SSCNN) system was developed to distinguish PDAC from Non-PDAC. The internal and external verifications were adopted and the diagnostic accuracy was compared between different seniorities of cytopathologists. 33 images of "Atypical" diagnosed by expert cytopathologists were selected to analyze the consistency between the system and definitive diagnosis.

RESULTS

The segmentation indicators Mean Intersection over Union (mIou), precision, recall and F1-score of SSCNN in internal and external testing sets were 88.3%, 93.21%,94.24%, 93.68% and 87.75%, 93.81%, 93.14%, 93.48% successively. The PDAC classification indicators of the SSCNN model including area under the ROC curve (AUC), accuracy, sensitivity, specificity, PPV and NPV in the internal testing set were 0.97%, 0.95%, 0.94%, 0.97%, 0.98%, 0.91% respectively, and 0.99%, 0.94%, 0.94%, 0.95%, 0.99%, 0.75% correspondingly in the external testing set. The diagnostic accuracy of senior, intermediate and junior cytopathologists was 95.00%, 88.33% and 76.67% under the binary diagnostic criteria of PDAC and non-PDAC. In comparison, the accuracy of the SSCNN system was 90.00% in the dataset of man-machine competition. The accuracy of the SSCNN model was highly consistent with senior cytopathologists (Kappa = 0.853, P = 0.001). The accuracy, sensitivity and specificity of the system in the classification of "atypical" cases were 78.79%, 84.20% and 71.43% respectively.

CONCLUSION

Not merely tremendous preparatory work was drastically reduced, the semi-supervised CNN model could effectively identify PDAC cell clusters in EUS-FNA cytological smears which achieved analogically diagnostic capability compared with senior cytopathologists, and showed outstanding performance in assisting to categorize "atypical" cases where manual diagnosis is controversial.

TRIAL REGISTRATION

This study was registered on clinicaltrials.gov, and its unique Protocol ID was PJ-2018-12-17.

Genome Mining for Hub Genes Related to Endoplasmic Reticulum Stress in Pancreatitis: A Perspective from In Silico Characterization

Pancreatitis, as a common exocrine pancreatic disease, poses a daunting challenge to patients' health and the medical system. Endoplasmic reticulum stress (ERS) plays an essential role in the pathologic process of pancreatitis. However, its mechanism is not fully understood. Therefore, this study was designed to deepen the understanding of the pathogenic mechanism of the disease by screening key ERS-related genes (ERSRGs) associated with pancreatitis. Pancreatitis mRNA data for GSE194331 (Normal: 32, Pancreatitis: 87) and pancreatitis GSE143754 (Normal: 9, Pancreatitis: 6) were downloaded from the GEO database and were used as a training and validation set, respectively. First, the training set GSE194331 was differentially expressed and intersected with the ERSRGs (n = 265) obtained from the MSigDB database to result in 42 differentially expressed ERSRGs (DE-ERSRGs). Subsequently, five candidate genes were further screened by PPI network and MCC and MCODE algorithms. However, according to the ROC curve results, AUC values of CCND1, BCL2, PIK3R1, and BCL2L1 were all greater than 0.6, indicating that they had good diagnostic performance, which was verified by the GSE143754 data set. Based on the GeneMANIA network, it was found that hub genes BCL2 and BCL2L1 may be the key factors in the regulation of mitochondrial metabolism. 24 differentially expressed pancreatitis-related genes (DE-PRGs) were found by difference analysis and Venn analysis. Hub genes BCL2 and PIK3R1 that were significantly correlated with 24 DE-PRGs were determined by Spearman analysis. ssGSEA algorithm was further used to reveal the significant correlation between these hub genes and the immune microenvironment of pancreatitis. The miRNA and lncRNA targeting hub genes were predicted using miRWalk, TargetScan, miRDB, and ENCORI databases, providing research directions for the mechanism of pancreatitis. Finally, the Network Analyst website was used to predict potential compounds associated with the hub gene. In conclusion, this study not only further supports the role of ERS in the development of pancreatitis but also provides a new perspective and direction for the development of biomarkers and mechanism of pancreatitis. At the same time, the results of this study provide a reliable research direction for the targeted treatment of pancreatitis.

Impact of Artificial Intelligence on Pancreaticobiliary Endoscopy

Pancreaticobiliary diseases can lead to significant morbidity and their diagnoses rely on imaging and endoscopy which are dependent on operator expertise. Artificial intelligence (AI) has seen a rapid uptake in the field of luminal endoscopy, such as polyp detection during colonoscopy. However, its use for pancreaticobiliary endoscopic modalities such as endoscopic ultrasound (EUS) and cholangioscopy remains scarce, with only few studies available. In this review, we delve into the current evidence, benefits, limitations, and future scope of AI technologies in pancreaticobiliary endoscopy.

A Semi-Supervised Learning Framework for Classifying Colorectal Neoplasia Based on the NICE Classification

Labelling medical images is an arduous and costly task that necessitates clinical expertise and large numbers of qualified images. Insufficient samples can lead to underfitting during training and poor performance of supervised learning models. In this study, we aim to develop a SimCLR-based semi-supervised learning framework to classify colorectal neoplasia based on the NICE classification. First, the proposed framework was trained under self-supervised learning using a large unlabelled dataset; subsequently, it was fine-tuned on a limited labelled dataset based on the NICE classification. The model was evaluated on an independent dataset and compared with models based on supervised transfer learning and endoscopists using accuracy, Matthew's correlation coefficient (MCC), and Cohen's kappa. Finally, Grad-CAM and t-SNE were applied to visualize the models' interpretations. A ResNet-backboned SimCLR model (accuracy of 0.908, MCC of 0.862, and Cohen's kappa of 0.896) outperformed supervised transfer learning-based models (means: 0.803, 0.698, and 0.742) and junior endoscopists (0.816, 0.724, and 0.863), while performing only slightly worse than senior endoscopists (0.916, 0.875, and 0.944). Moreover, t-SNE showed a better clustering of ternary samples through self-supervised learning in SimCLR than through supervised transfer learning. Compared with traditional supervised learning, semi-supervised learning enables deep learning models to achieve improved performance with limited labelled endoscopic images.

Radiogenomic analysis for predicting lymph node metastasis and molecular annotation of radiomic features in pancreatic cancer

BACKGROUND

To provide a preoperative prediction model for lymph node metastasis in pancreatic cancer patients and provide molecular information of key radiomic features.

METHODS

Two cohorts comprising 151 and 54 pancreatic cancer patients were included in the analysis. Radiomic features from the tumor region of interests were extracted by using PyRadiomics software. We used a framework that incorporated 10 machine learning algorithms and generated 77 combinations to construct radiomics-based models for lymph node metastasis prediction. Weighted gene coexpression network analysis (WGCNA) was subsequently performed to determine the relationships between gene expression levels and radiomic features. Molecular pathways enrichment analysis was performed to uncover the underlying molecular features.

RESULTS

Patients in the in-house cohort (mean age, 61.3 years ± 9.6 [SD]; 91 men [60%]) were separated into training (n = 105, 70%) and validation (n = 46, 30%) cohorts. A total of 1,239 features were extracted and subjected to machine learning algorithms. The 77 radiomic models showed moderate performance for predicting lymph node metastasis, and the combination of the StepGBM and Enet algorithms had the best performance in the training (AUC = 0.84, 95% CI = 0.77-0.91) and validation (AUC = 0.85, 95% CI = 0.73-0.98) cohorts. We determined that 15 features were core variables for lymph node metastasis. Proliferation-related processes may respond to the main molecular alterations underlying these features.

CONCLUSIONS

Machine learning-based radiomics could predict the status of lymph node metastasis in pancreatic cancer, which is associated with proliferation-related alterations.

Application of artificial intelligence in pancreas endoscopic ultrasound imaging- A systematic review

The pancreas is a vital organ in digestive system which has significant health implications. It is imperative to evaluate and identify malignant pancreatic lesions promptly in light of the high mortality rate linked to such malignancies. Endoscopic Ultrasound (EUS) is a non-invasive precise technique to detect pancreas disorders, but it is highly operator dependent. Artificial intelligence (AI), including traditional machine learning (ML) and deep learning (DL) techniques can play a pivotal role to enhancing the performance of EUS regardless of operator. AI performs a critical function in the detection, classification, and segmentation of medical images. The utilization of AI-assisted systems has improved the accuracy and productivity of pancreatic analysis, including the detection of diverse pancreatic disorders (e.g., pancreatitis, masses, and cysts) as well as landmarks and parenchyma. This systematic review examines the rapidly developing domain of AI-assisted system in EUS of the pancreas. Its objective is to present a thorough study of the present research status and developments in this area. This paper explores the significant challenges of AI-assisted system in pancreas EUS imaging, highlights the potential of AI techniques in addressing these challenges, and suggests the scope for future research in domain of AI-assisted EUS systems.

Validation of a real-time biliopancreatic endoscopic ultrasonography analytical device in China: a prospective, single-centre, randomised, controlled trial

BACKGROUND

Endoscopic ultrasonography (EUS) is a key procedure for the diagnosis of biliopancreatic diseases. However, the performance among EUS endoscopists varies greatly and leads to blind spots during the operation, which can impair the health outcomes of patients. We previously developed an artificial intelligence (AI) device that accurately identified EUS standard stations and significantly reduced the difficulty of ultrasonography image interpretation. In this study, we updated the device (named EUS-IREAD) and validated its performance in improving the quality of EUS procedures.

METHODS

In this single-centre, randomised, controlled trial, we updated EUS-IREAD so it consisted of five learning models to identify eight EUS stations and 24 anatomical structures. The trial was done at the Renmin Hospital of Wuhan University (Wuhan, China) and included patients aged 18 years or older with suspected biliopancreatic (pancreas and biliopancreatic duct) lesions due to clinical symptoms, radiological findings, or laboratory findings, and with a high risk of pancreatic cancer. Patients were randomly assigned (1:1) by a dedicated research assistant using a computer-generated random number series (with a block size of four) to undergo the EUS procedure with or without the assistance of EUS-IREAD. Endoscopists in the EUS-IREAD-assisted group were required to observe all standard stations and anatomical structures according to the prompts by the AI device. Data collectors, the independent data anaylsis team, and patients were masked to group allocation. The primary outcome was the missed scanning rate of standard stations between the two groups, which was assessed in patients who underwent EUS procedure in accordance with the assigned intervention (per protocol). This trial is registered with ClinicalTrials.gov, NCT05457101.

FINDINGS

Between July 9, 2022, and Feb 28, 2023, 290 patients (mean age 55·93 years [SD 14·06], 152 [52%] male, and 138 [48%] female) were randomly assigned and analysed, including 144 in the EUS-IREAD-assisted group and 146 in the control group. The EUS-IREAD-assisted group had a lower missed scanning rate of stations than the control group (4·5% [SD 0·8] vs 14·3% [1·0], -9·8% [95% CI -12·2 to -7·5]; odds ratio 3·6 [95% Cl 2·6 to 4·9]; p<0·0001). No significant adverse event was found during the study.

INTERPRETATION

Our study confirms the capability of EUS-IREAD to monitor the blind spots and reduce the missed rate of stations and structures during EUS procedures. The EUS-IREAD has the potential to play an essential part in EUS quality control.

FUNDING

Innovation Team Project of Health Commission of Hubei Province and College-enterprise Deepening Reform Project of Wuhan University.

Artificial Intelligence in Endoscopic Ultrasonography-Guided Fine-Needle Aspiration/Biopsy (EUS-FNA/B) for Solid Pancreatic Lesions: Opportunities and Challenges

Solid pancreatic lesions (SPLs) encompass a variety of benign and malignant diseases and accurate diagnosis is crucial for guiding appropriate treatment decisions. Endoscopic ultrasonography-guided fine-needle aspiration/biopsy (EUS-FNA/B) serves as a front-line diagnostic tool for pancreatic mass lesions and is widely used in clinical practice. Artificial intelligence (AI) is a mathematical technique that automates the learning and recognition of data patterns. Its strong self-learning ability and unbiased nature have led to its gradual adoption in the medical field. In this paper, we describe the fundamentals of AI and provide a summary of reports on AI in EUS-FNA/B to help endoscopists understand and realize its potential in improving pathological diagnosis and guiding targeted EUS-FNA/B. However, AI models have limitations and shortages that need to be addressed before clinical use. Furthermore, as most AI studies are retrospective, large-scale prospective clinical trials are necessary to evaluate their clinical usefulness accurately. Although AI in EUS-FNA/B is still in its infancy, the constant input of clinical data and the advancements in computer technology are expected to make computer-aided diagnosis and treatment more feasible.