Introduction

Even if early gastric cancer is detected endoscopically, if a pathological diagnosis of cancer is not made based on biopsy tissue, there is a risk that curative endoscopic resection will not be performed, and the cancer will be left untreated until it becomes an advanced cancer. The intramucosal component of early gastric cancer collected by biopsy often shows no signs of invasion and has low atypia, making it often difficult to determine whether it is a neoplasia or a non-neoplasia. To solve these problems, we developed an e-learning system that improves the pathological diagnosis of gastric biopsies and evaluated its effectiveness using an Internet connection.

Aims & Methods

The design of this study was a randomized controlled trial. All tests and e-learning were conducted using the Internet. After receiving the pre-test (60 questions on biopsy tissue photos), participants were randomly assigned to either an e-learning group (E-group) or a non-e-learning group (NE-group), and only members of the E-group had access to the e-learning system. Two months later after the pre-test, both two groups received the post-test (the same 60 questions as the pre-test). The primary endpoint was the difference in the rate of improvement in test results between the two groups.

Results

215 pathologists from 36 countries who were assessed for eligibility were included in this study, with 106 assigned to the E-group and 109 to the NE-group. Using the test score (88.3 points) of the top 30% (34/109) of the NE-group as the cutoff value, high-ranking participants (more than 88.3 points) were 57 among 106 participants in the E-group. There were significantly more high-ranking participants in the E-group than in the NE-group (p=0.00048).

Conclusion

This global research demonstrated the efficacy of the developed e-learning system in improving the histopathological diagnosis of gastric cancer.

Introduction

Gastric cancer has been identified as a serious global public health problem, and early detection of gastric cancer is an effective means of curbing the risk of gastric cancer. However, accurate, efficient and economical detection of early gastric cancer (EGC) remains a challenge.

Aims & Methods

We aim to fully explore the characteristics of various endoscopic images and EGC diagnostic requirements to boost the accuracy and efficiency of deep learning-based EGC detection and enable its public availability, so as to improve the survival rate and life quality of patients. A new lightweight EGC fusion diagnosis framework based on heterogeneous neural networks (FusionWN-Net) is proposed. FusionWN-Net consists of two stages: Model construction and Decision-level fusion. In the Model construction, two lightweight EGC classification models (WLI-Net and NBI-Net) are designed for the imaging characteristics of WLI and NBI to obtain accurate unimodal diagnostic results while guaranteeing low computational costs; in the Decision-level fusion, the Dempster-Shaffer algorithm is utilized to further fuse the diagnostic results of WLI-Net and NBI-Net to reach the final diagnosis. The proposed framework is trained and evaluated on a dataset of gastric endoscopic images collected and established from 1,055 patients of six hospitals, including Fuzhou University Affiliated Provincial Hospital. On this basis, we use Android Studio to develop a portable EGC diagnosis system (PEGCS) which can run smoothly on the mobile device.

Results

Extensive experiments show that the proposed framework exhibits superior performance in terms of both computational efficiency and diagnostic reliability. WLI-Net achieves 83.31% diagnostic accuracy with only 3.182M parameters and 0.471G FLOPs, and 37.04 FPS on the mobile device; NBI-Net achieves 82.85% diagnostic accuracy with 3.596M parameters and 0.609G FLOPs, and 24.39 FPS on the mobile device; the further obtained FusionWN-Net increases the accuracy to 86.52% with 6.778M parameters and 1.080G FLOPs, reaching the FPS of 14.08 on the mobile device, which is expected to be used in clinical practice.

Conclusion

We proposed WLI-Net and NBI-Net that can be used for white light endoscopic images and narrow band endoscopic images respectively, and a lightweight and efficient multimodal framework for EGC diagnosis, which are tailored for the unique characteristics and diagnostic requirements of different endoscopic images. Besides, we established the PEGCS to put the framework into practice. This work not only takes a big step forward in improving the EGC diagnostic accuracy, but also provides a general and efficient solution for obtaining high-quality EGC diagnosis results in resource-limited application scenarios, which demonstrates significant potential of clinical application.

Disclosure

The authors declare no conflicts of interest, financial or otherwise, related to the preparation or publication of this work.

Introduction

Endoscopic diagnosis of gastric dysplasia (GD) and early gastric cancer (EGC) remains challenging. Raman spectral analysis is a biomolecular specific vibrational analytical technique which measures inelastic light scattering process within samples. This technique detects disease via biomolecular "fingerprinting", which may be applied to objectively diagnose gastric lesions.

Aims & Methods

We aim to develop a machine learning model based on Raman spectra to diagnose intestinal metaplasia (IM), GD and EGC with biomarker correlations. Raman spectra were prospectively collected prior to endoscopic resection and correlated with histopathology. GD and EGC were classified as high-risk (HR) lesions; while IM and gastritis as low-risk (LR). Two methods were employed to identify significant Raman wavenumber for group differentiation: Partial Least Squares-Discriminant Analysis with Variance Importance in Projection analysis, and a theory-based statistical approach. Next, these features were tested using XGBoost model for best accuracy. A similar approach was applied to differentiate IM versus LGD, and the model was tested with leave-one-out cross-validation (LOOCV) method for the Support Vector Machine (SVM).

Results

20 participants were recruited from April 2022 to July 2023. A total of 1961 Raman spectra were collected from 33 lesions for HR versus LR differentiation. The results are as shown in Table 1. The diagnostic accuracy and F1 score for PLS-DA VIP and theory-based ranking were similar at 75.5% and 0.8318, respectively. XGBoost testing revealed that the model attained peak performance with the top 400 wavenumbers from PLS-DA VIP sorting, and the top 200 wavenumbers from the theory-based ranking. An overlap of 192 wavenumbers emerged, and the model achieved a peak F1 Score of 0.84 and 76.9% accuracy. These 192 wavenumbers were then mapped to the Raman spectrum and can be correlated with 8 Raman peaks of biochemicals related to gastric cancer including lipids, proteins, phospholipids and leucine. For LGD versus IM, we utilized 487 Raman signals from 25 lesions. was an overlap of 100 wavenumbers with accuracy of 74.1%.
The diagnostic accuracy and F1 score for PLS-DA VIP and theory-based ranking were similar at 74.4% and 0.79 respectively. LOOCV with SVM testing revealed that the model attained peak performance with the top 200 wavenumbers from PLS-DA VIP sorting, and the top 100 wavenumbers from the theory-based ranking. An overlap of 100 wavenumbers emerged, and the model achieved a peak F1 Score of 0.80 and 74.7% accuracy.

Table 1 Current AI model (XGBoost for differentiaton high risk and low risk lesions; LOOC method with SVM performance: intestinal metaplasia ) performance with selected wavenumbers for differentiation between gastric lesions.

	Number of wavenumbers	Accuracy	F1-Score
High-risk versus low-risk
Machine learning approach	400	75.5%	0.8318
Statistical approach	200	75.5%	0.8318
Pivotal wavenumbers	192	76.9% ↑	0.8426 ↑
Low grade dysplasia versus intestinal metaplasia
Machine learning approach	200	74.7%	0.7876
Statistical approach	100	74.7%	0.796
Pivotal wavenumbers	100	74.7%	0.796

Abbreviations: LOOC: leave-one-out cross-validation; SVM: Support Vector Machine

Conclusion

A Raman spectroscopy-based AI algorithm may be useful for real time diagnosis of GIM, GD and EGC.

Introduction

Gastric cancer (GC) ranks as the sixth most common cancer globally and is the fifth leading cause of cancer-related mortality. Previous studies have shown that the five-year survival rate for early-stage GC exceeds 90%. Gastroscopy is the gold standard for GC diagnosis, but its invasive nature and high cost limit its clinical application. Non-invasive tests have been reported in GC prediction, yet current biomarkers fall short in sensitivity or specificity, particularly for early GC prediction, leading to misdiagnosis and potential delays in GC treatment, which may reduce patient survival rates. Multi-omics biomarker panels significantly enhance diagnostic efficacy through multidimensional information complementarity, compared to single-omics or individual biomarkers. Therefore, developing non-invasive multi-omics biomarker combinations holds great promise for enhancing GC diagnostic accuracy and increasing the likelihood of early diagnosis and treatment.

Aims & Methods

This study utilized mass spectrometry-based proteomics and metabolomics to comprehensively analyze the peripheral blood components of non-gastric cancer (NGC, including patients with chronic superficial gastritis and precancerous lesions such as atrophic gastritis, intestinal metaplasia, and dysplasia) and gastric cancer (GC, including patients with well-defined staging from stage IA (T1N0M0) to advanced stages). We employed machine learning algorithms to develop a multi-omics biomarker combination capable of identifying GC and enabling early GC diagnosis.The studies involving humans were approved by the Medical Research Ethics Committee of Peking University Third Hospital.

Results

Through differential analysis of plasma proteins and metabolites in GC (n=53) and NGC (n=41) (FDR<0.05, log2FC>1), we identified 333 upregulated proteins and 570 upregulated metabolites. LASSO regression further narrowed down the candidates to 10 proteins (10-P) and 10 metabolites (10-M) as potential biomarkers. Subsequently, we constructed gastric cancer prediction models using random forest algorithms. In the validation set (n=27), the 10-P Model achieved a prediction accuracy of 0.85 (AUC: 0.91, sensitivity 0.83, specificity 0.86), while the 10-M Model had a prediction accuracy of 0.77 (AUC: 0.82, sensitivity 0.83, specificity 0.77). The Multi-Omics Integrated Model (MOIM), built via machine learning-based Voting Classifier, reached a prediction accuracy of 0.92 (AUC: 0.96, sensitivity 0.91, specificity 0.93), surpassing the former two models. For stage IA GC (n=22) prediction, MOIM's accuracy was 0.93, higher than the 10-P Model's 0.86 and the 10-M Model's 0.79. Finally, we streamlined the model to include only 7 proteins and 3 metabolites, forming the Simplified Multi-Omics Integrated Model (sMOIM), which achieved accuracies of 0.91 for GC and 0.92 for stage IA GC.

	GC(N=53) VS NGC(N=41)				stage IA GC(N=22)
	Accuracy	AUROC	Sensitivity	Specificity	Accuracy	AUROC	Sensitivity	Specificity
10-P model	0.85	0.91	0.83	0.86	0.86	0.92	0.91	0.73
10-M model	0.77	0.82	0.83	0.77	0.79	0.85	0.83	0.79
MOIM	0.92	0.96	0.91	0.93	0.93	0.98	0.95	0.90
sMOIM	0.91	0.95	0.92	0.91	0.92	0.96	0.93	0.91

Conclusion

The selected plasma protein and metabolite biomarker combination enables highly accurate prediction of GC, especially early-stage GC. Moreover, the multi-omics integrated analysis achieves superior accuracy, sensitivity, and specificity compared to single-omics prediction, suggesting its potential as an effective strategy for the precise early diagnosis of gastric cancer.

Introduction

Recently, artificial intelligence (AI)-based diagnostic support tools for early gastric cancer have been developed and reported to be superior to endoscopists (accuracy: 70% and sensitivity: 85%).¹ However, These tools could not guarantee reliable diagnostic accuracy in clinical practice. In previous reports, magnifying endoscopy has been recognized as superior to white-light imaging endoscopy for endoscopic diagnosis of early gastric cancers. Therefore, we developed an AI-based diagnostic support system using magnifying endoscopy with narrow band imaging (M-NBI) for early gastric cancers² and improved its application in clinical practice using nondegraded images obtained from selective depth inspection (SDI) output for real-time diagnosis. This study is the first to create an AI-based diagnostic support system using non-degraded M-NBI images obtained from SDI output, as several previous papers have analyzed damaged M-NBI images that were converted from various recording devices.^3,4

Aims & Methods

We built a new AI system (M-NBI AI) which was assembled using four different image recognition models with non-degraded M-NBI images (normal mucosa:1776 images, gastric cancers: 1947 images, and non-neoplastic lesions: 2451 images). We evaluated the performance of this M-NBI AI for diagnosis of early gastric cancer. The diagnostic performances of M-NBI AI and endoscopists (five specialists and five non-specialists) were compared using non-degraded M-NBI images (three images per lesion) of 50 early gastric cancer and 50 non-cancerous lesions. Early gastric cancers included histopathological types of well, moderately, and poorly differentiated adenocarcinomas and signet-ring cell adenocarcinomas. The primary outcome was the superiority of the M-NBI AI (based on accuracy) over the endoscopists. The secondary outcome was the superiority of the M-NBI AI (based on sensitivity) over the endoscopists.

Results

	M-NBI AI	Endoscopists
		Specialists	Non-specialists
Accuracy, % (95% CI)	87.0 (0.80–0.94)	80.0 (0.72–0.88)	71.0 (0.62–0.79)
Sensitivity, % (95% CI)	86.3 (0.77–0.96)	74.2 (0.63–0.85)	70.6 (0.58–0.83)
Specificity, % (95% CI)	88.0 (0.79–0.97)	89.0 (0.80–0.99)	71.0 (0.59–0.84)
PPV, % (95% CI)	88.0 (0.79–0.97)	92.0 (0.84–0.99)	72.0 (0.60–0.84)
NPV, % (95% CI)	86.0 (0.76–0.96)	68.0 (0.55–0.81)	70.0 (0.57–0.83)
PPV, positive predictive value; NPV, negative predictive value

The M-NBI AI was superior to the non-specialists based on accuracy (87.0%, 95%CI: 0.80–0.94 versus 71.0 %, 95%CI: 0.62–0.79; p<0.01), without being superior to the specialists (87.0%, 95%CI: 0.80–0.94 versus 80.0%, 95%CI: 0.72–0.88; p=0.18). The M-NBI AI was superior to the specialists and non-specialists based on sensitivity (86.3%, 95%CI: 0.77–0.96 versus 74.2%, 95%CI: 0.63–0.85 and 70.6%, 95%CI: 0.58–0.83; p<0.05, respectively).

Conclusion

The M-NBI AI was created as a new AI system that could be implemented for current endoscopy systems as a real-time diagnostic support tool for early gastric cancers. It demonstrated a robust performance and may indicate equalization of diagnostic accuracy and superadded effect over the white-light endoscopy for diagnosis of early gastric cancers in clinical practice.

References

1. Ishioka M et al. Dig Endosc. 2022. doi: 10.1111/den.14455.
2. Ueyama H et al. J Gastroenterol Hepatol. 2021;36(2):482-489.
3. Horiuchi Y et al. Dig. Dis. Sci. 2020;65:1355–1363.
4. He X et al. Gastrointest Endosc. 2022;95(4):671-678.

Introduction

Gastric cancer (GC) is a major cause of cancer-related death worldwide. Although immune checkpoint inhibitors (ICIs) have improved outcomes in a subset of patients, only 20–30% achieve long-term benefit. Identifying robust biomarkers to predict immunotherapy response remains a critical need. With the rise of digital pathology and deep learning (DL), it is now possible to derive prognostic features from whole-slide images (WSIs), offering a novel means of evaluating the tumor immune microenvironment.

Aims & Methods

This study aimed to develop a Digital Pathology-based Immunotherapy Response Signature (DPIRS) using DL to predict prognosis in GC patients treated with ICIs. We retrospectively analyzed 110 GC patients receiving ICIs at The First Hospital of Jilin University between 2019–2023. H&E-stained WSIs were segmented into 256×256 pixel tiles. A convolutional neural network (ResNet18) was trained to detect immune-relevant histopathological patterns. The DPIRS score for each patient was defined as the proportion of positively scored image tiles. DPIRS was correlated with progression-free survival (PFS), overall survival (OS), immune checkpoint molecule expression, and immune infiltration characteristics.

Results

The DL model showed excellent predictive performance (AUC >0.85) across internal and TCGA validation cohorts. Patients with high DPIRS scores had significantly longer PFS and OS (p<0.01), as well as higher PD-L1 expression and greater tumor-infiltrating lymphocyte (TIL) density. Multivariate Cox regression confirmed DPIRS as an independent prognostic factor for immunotherapy efficacy.

Conclusion

DPIRS is a novel, non-invasive biomarker derived from deep learning-based analysis of digital pathology, enabling precise prediction of immunotherapy response in gastric cancer. This tool holds strong potential for clinical application in personalized treatment strategies.