Artificial intelligence applications in computed tomography in gastric cancer: a narrative review

Introduction

Gastric cancer (GC) is the fifth most common malignancy and the fourth reason for cancer-related deaths (1). Furthermore, in several South-Central Asian countries, GC is the most common and fatal cancer in men (1). Attributed to the endoscopic screening, the mortality of GC decreased in some countries. However, the overall 5-year survival rate remains poor due to asymptomatic onset and delayed diagnosis (2).

Artificial intelligence (AI) is a revolutionary technique which is deeply impacting and reshaping clinical practice in oncology (3). It is an umbrella term that describes computing techniques for simulating human intelligence (4). Machine learning (ML) is a domain branch of AI. It is defined as computing algorithms that enable self-learning of input data patterns without explicit instruction (5). Deep learning (DL) is an important subset of ML. DL is defined as a learning algorithm that can automatically learn unknown features, maximizing classification with limited supervision through multiple layers of artificial neural network (6,7). As an advanced ML method, DL has received widespread attention and made breakthroughs in various fields such as computer vision tasks and clinical applications by simulating the human brain for analysis, learning, and data analysis. DL performs representation learning on data, using multiple processing layers composed of multiple non-linear transformations to achieve high-level abstractions of the data (8,9). The biggest characteristic of DL is its ability to autonomously learn through multi-layer neural networks, approximating complex functions through deep non-linear network structures and directly obtaining features associated with the data. DL has revolutionized the algorithmic design approach in many domains including speech recognition, image classification, and text understanding, gradually forming a new paradigm that uses end-to-end models to directly output final results from training data (10). Compared to traditional data feature extraction processes, DL reduces the need for manual preprocessing steps and enables autonomous ML and feature extraction (8). The features obtained by DL thus provide a more fundamental characterization of the data and exhibit significant advantages in terms of classification and visualization. Additionally, due to the robustness of DL, it can typically avoid unexpected variables that researchers may not anticipate, such as inter-observer variability and different clinical conditions and scanning parameters (11,12). With the advancement of computing power and graphic processing technologies, AI techniques such as segmentation, detection and classification are being increasingly utilized to the field of medical imaging (11,13,14). AI has demonstrated the potential to outperform clinicians in diagnostic accuracy and time-saving (14).

Although endoscopy is the most effective tool for early detection of GC, it cannot identify metastatic lesions, which may lead to mismanagement of patients (15). The diagnostic accuracy is mainly dependent on the experience of endoscopists (16). Biopsy specimens and pathological slices are difficult to capture the heterogeneity over the whole tumor. Therefore, computed tomography (CT) is currently the most commonly used and convenient tool for diagnosing, preoperative staging, and treatment efficacy evaluation for GC (17). However, the consistency and accuracy of image interpretation vary largely, which may not improve with training and experience (9). The analysis of CT images mainly relies on morphological features, which provide limited information underlying tumor development and progression (18). These problems present inevitable challenges for radiologists in personalized and precision medicine. Due to the desire to solve these problems, AI applications in CT images attracted considerable attention in various fields such as image segmentation, diagnosis, prediction of metastasis risk, survival, and treatment response (9). AI techniques provide new methods to process images and translate them into quantitative data, allowing the identification of microscopic features of tumors invisible to human eyes (14). Nevertheless, promising results of AI-assisted CT analysis have been reported. Several concerns have also accompanied the surge of AI applications in clinical practice.

In this review, we will summarize the current status of the clinical application of AI-based CT for GC, focusing on diagnosis, genetic status detection and risk prediction of metastasis, prognosis and treatment efficacy. The challenges and prospects for future research will also be discussed. We present this article in accordance with the Narrative Review reporting checklist (available at https://tcr.amegroups.com/article/view/10.21037/tcr-23-201/rc).

Methods

We searched the PubMed database to identify studies investigating AI applications in CT in GC. The latest search time was 1/11/2022. The search terms used were: “gastric cancer” and (“artificial intelligence” or “machine learning” or “deep learning” or “radiomics”) and “computed tomography”. The secondary references cited in articles obtained from the MEDLINE and PubMed search were also retrieved. We only considered research articles written in English. Two authors read the potential studies and wrote the draft. The search strategy is summarized in Table 1. The characteristics of key studies are summarized in Table 2.

Diagnosis and differential diagnosis

Accurate T-staging is essential for selecting appropriate patients for neoadjuvant chemotherapy (NAC) (55). However, 22.8% of T1/2 GC were misdiagnosed as stage T3/4 by a combination of endoscopy and CT preoperatively (56). Wang and colleagues reported a CT-based radiomics model accuracy for differentiation of T2 and T3/4 stage GC, with an optimal area under the curve (AUC) of 0.899 (19). Sun et al. extracted handcrafted DL features from three phase contrast-enhanced CT (CECT) and built a radiomics signature using ML methods (20). Unlike traditional radiomics analysis, this study did not only used handcraft features, but also features extracted from DL neural networks, which provided more diagnostic information beyond a visual interpretation. Their radiomics nomogram, combined with radiomics signature and conventional CT signs, performed well in identifying T4a stage GC, with an optimal AUC of 0.90 (20).

Primary gastric lymphoma (PGL) is one of the most diagnosed gastric malignancies, difficult to differentiate from Borrmann type IV GC using conventional CT (21). The treatment options for the two diseases are different (57). Surgery is the main treatment option for GC, while chemo and radiotherapy are the best options for PGL (57). Ma et al. reported a radiomics based model differentiating Borrmann type IV GC from PGL. The model showed favorable accuracy with an AUC of 0.903 (21). DL algorithms typically require a substantial amount of data for effective training. Therefore, this requirement often leads to the challenge of data scarcity under ideal circumstances (58). Transfer learning is an ML technique that enhances the performance of a target task by leveraging knowledge acquired from a related source task. This approach enables the reduction of data requirements for the target task, making it less reliant on a large amount of data (59). Feng et al. proposed a transfer learning model based on CT and whole slide images. The model achieves high accuracy for distinguishing PGL from Borrmann type IV GC, with AUCs ranging from 0.92 to 0.99 (22).

Recently, Wang and colleagues constructed a radiomics nomogram with radiomics signature, lymph node metastasis (LNM), and tumor margin. This nomogram revealed a moderate accuracy for differentiating gastric neuroendocrine carcinomas from adenocarcinomas (23). Chen et al. reported a radiomics-based nomogram for differential diagnosis for diffuse-type and signet ring cell GC (24).

Risk prediction of metastasis

LNM prediction

LNM is one of the most important hazard factors for the unfavorable outcome of GC (60). Although CT is the most commonly used tool for preoperative assessment of LNM, conventional CT mostly relies on morphological characteristics with unsatisfactory sensitivity for LNM (61). Huang et al. firstly revealed the value of radiomics for predicting LNM in colorectal cancer (62). In this study, 150 radiomics features were extracted from CT images and reduced to 24 optimal predictors by using the least absolute shrinkage and selection operator (LASSO) model with 10-fold cross-validation. Then, a radiomics signature was established with linear combination of selected features. The radiomics signature demonstrated significant association with LNM status. Since then, radiomics and DL approaches have attracted great interest for LNM prediction in GC (15,25,26,28,63,64). Several studies developed radiomic models by extracting features from CECT images and found promising results (15,25,63,65,66). Dong et al. enrolled 730 locally advanced GC (AGC) patients from five independent centers and proposed a DL radiomic nomogram based on multiphase CT. The nomogram showed good predictive ability for N-staging with best AUC of 0.822 (25). Dual-energy CT (DECT) provides quantitative information on material concentration and improves the ability of tumor characterization. Li et al. extracted DL and radiomics features from three monochromatic images for arterial phase (AP) or portal venous phase (VP) image of CECT (27). They found that AP and VP radiomics showed significant association with LNM. The nomogram incorporated the two radiomics signatures, and CT-reported LNM achieved better performance in both cohorts with AUCs of 0.84 and 0.82 (27). For GC confined to the mucosa and submucosa, patients without LNM can receive endoscopic resection in order to avoid complication of lymphadenectomy. Nevertheless, up to 25% of T1b (submucosa) patients suffer LNM. Endoscopic resection for EGC with LNM may delay the detection of LNM and prevent early management of patients. Thus, accurate preoperative diagnosis of LNM is of particular significance for patients with early-stage GC. Gao et al. proposed a radiomics signature with 6 features and revealed favorable accuracy for LNM prediction for early-stage GC (17).

While most LNM prediction models were based on images of primary gastric lesions, Gao et al. proposed a faster region-based convolutional neural network (CNN) model by labeling perigastric metastatic lymph nodes. The recognition accuracy of their model achieved 95.4% (64). Jin et al. proposed a CNN-based model for predicting 11 lymph node stations. The model showed excellent prediction accuracy with an AUC of 0.88. Furthermore, by visualizing subnetworks, they found that imaging features related to intratumoral heterogeneity and invasive seem to be most valuable for predicting LNM (28).

Lymphovascular invasion (LVI)

LVI, which refers to lymphatic and/or blood vessel invasion, is an independent risk factor for LNM and early recurrence in GC patients (67,68). Chen et al. developed a radiomics signature based on AP and VP images to predict LVI (18). The radiomics signature combined with AP and VP features achieved good predictive accuracy with an optimal AUC of 0.865. Fan et al. constructed a radiomics model integrating CECT- and PET-based features for predicting LVI in GC. Three ML classifiers (adaptive boosting, linear discriminant analysis, and logistic regression) were used for model construction. A combined model with 10 radiomics features and 8 clinical factors showed the best performance with AUCs of 0.92–0.94 (29). Li et al. developed two LVI prediction models for GC using radiomics and the transfer learning method (69). The two models showed similar prediction performances. They further constructed a nomogram incorporating the radiomics model, histologic grade, radiological T, and N stage. The nomogram also revealed good predictive ability for progression-free survival (PFS) and overall survival (OS).

Peritoneal metastasis (PM)

The peritoneum is one of the most common sites of metastasis in GC (70). However, occult PM, which shows no positive signs on CT images, occurs in 10–30% of GC patients (71). Therefore, tools for detecting occult PM are an urgent need. Several studies have evaluated the value of CT-based ML methods for predicting occult PM (30-34). In contrast to conventional radiomics workflow, which only extracts features from the primary tumor area, Dong et al. built two radiomics signatures by features extracted from the tumor area and peritoneal area. Their results showed that radiomics features from both areas could provide useful information on PM. The nomogram incorporated the radiomics signatures and Lauren types achieved an extremely high accuracy for predicting PM with (31). In another study conducted by Chen and his colleagues, iodine uptake (IU) and 120-kV equivalent mixed (M) images were acquired for feature extraction by using DECT (34). They found that IU-derived radiomics features provide useful information for predicting PM. Since adjacent nontumor tissues may also provide a wealth of information of tumorigenesis, Liu and colleagues applied the bounding box annotation method, which contained both the tumor and peritumoral area, to develop a radiomics-based model for the prediction of PM (35). Unlike traditional label methods of lesions, bounding box method can encompass both the tumor and peritumoral area. This allows for the inclusion of more information compared to traditional annotation methods, making it more convenient and time-saving (72,73).

Huang et al. first investigated the value of DL model for identifying occult PM for AGC. Their proposed deep CNN model based on 2D CECT images revealed promising accuracy with an AUC of 0.90 (36). In a multicenter study that enrolled 1,978 GC patients, Jiang developed a deep CNN model with long-short connections for predicting occult PM (37). In contrast to a conventional CNN, their model incorporates a long connection that facilitates the extraction of multilevel tumor features. These features are then integrated into the final fully connected layer for prediction (37). Their model achieved better performance with best AUCs of 0.946.

Genetic status and molecular subtypes

With the development of high-throughput sequencing techniques, several genomic classification systems reflecting the complicated genomic mechanisms underlying GC have been proposed. The Cancer Genome Atlas (TCGA) has proposed a molecular classification including four subgroups: Epstein-Barr virus (EBV) status, microsatellite instable, genome stable and chromosomal instability. This system has been incorporated into the World Health Organization (WHO) classification (74,75). Epithelial-to-mesenchymal transition and TP53 mutation are key players in the development and progression of tumor. Based on these two mechanisms, the Asian Cancer Research Group categorized GC into four molecular subtypes, which correlate well with the progression and outcome of GC (76). The molecular subtypes showed better stratification ability for adjuvant chemotherapy (AC) and immunotherapy response (77,78). Thus, preoperative identification of molecular subtypes is of great significance for treatment management for GC patients. However, researches focusing on the association between AI methods and genetic information of GC are still rare. EBV-positive GC is related to better survival and response of chemotherapy. By utilizing the public database of TCGA and The Cancer Imaging Archive, Zhao and colleagues constructed two radiomics models based on 2D or 3D features (38). Both models showed a good discriminating ability of EBV expression status. Zhang et al. developed a 9-feature radiomics model for predicting EBV status (39). The model also showed favorable accuracy with a sensitivity of 80% and a specificity of 84%.

The overexpression of human epidermal growth factor 2 (HER2) associates with a poor clinical outcome and plays a crucial role in tumorigenesis in GC (79,80). The ToGA trial demonstrated that anti-HER2 monoclonal antibody trastuzumab with chemotherapy prolonged survival of HER2⁺ AGC patients (81). Therefore, trastuzumab is recommended as the first-line treatment for AGC patients with HER2 overexpression by National Comprehensive Cancer Network guidelines. However, GC is a highly heterogeneous disease. In contrast to breast cancer, which generally exhibits homogenous HER2 overexpression, intratumoral heterogeneity of HER2 status has been observed in GC (82,83). Therefore, an accurate evaluation of HER2 status is of great importance for predicting the treatment efficacy of HER2 target therapy. Wang et al. developed two radiomics models based on the AP and VP images for HER2 expression status in GC (40). Both models showed moderate accuracy for discriminating HER2-negative GC with AUCs ranging from 0.72 to 0.83.

Prognosis prediction

As GC is highly heterogeneous with significant clinical characteristics variations even within the same stage, the commonly used tumor-node-metastasis (TNM) staging system offers insufficient information for outcome prediction. Moreover, the TNM stage can only be precisely confirmed postoperatively. Therefore, several studies have investigated the potential and their added value of radiomics- and DL-based approaches for prediction of survival and recurrence risk of GC (41-44).

Li et al. found that radiomics features could effectively predict the OS of GC patients undergone radical resection (41). Jiang and colleagues built a radiomics signature with 19 features related to disease-free survival (DFS) and OS (42). Jin et al. proposed a radiomics signature for predicting DFS and OS for GC patients. In addition, the radiogenomics analysis showed that some of the radiomics features might be associated with genes involved in drug metabolisms and chemokine regulation (43). Shin et al. have explored the association between CT-based radiomics features and reccurence-free survival (RFS) in 420 locally AGC patients. A merged model consisting of the radiomics signature and significant clinical parameters showed better accuracy than the radiomics signatures and clinical parameters alone (44).

Jiang et al. utilized a novel deep neural network named “S-net” that integrated comprehensive multi-scale image features to predict the DFS and OS of GC patients (45). Unlike conventional CNN architectures, the S-net model integrates the idea of multi-level feature stream fusion. This design choice is based on the understanding that both low-level features from shallow layers and high-level features from deep layers contain valuable information for survival analysis. By incorporating this approach, the S-net model can effectively extract and integrate comprehensive multi-scale image features, enabling a thorough understanding of complex tumor phenotypes (45). While S-net leverages various high-level features for survival prediction (45), it overlooks the valuable shallow information due to the repeated pooling and convolution operations. Therefore, Zhang et al. further proposed a multi-focus fusion feature pyramid network which unified separate lower-level and fused higher-level features. In this model, a new designed strategy of cascade connection which extracts single and fused lower-level features maps in shadow bottom-top pathway. The model showed better performance than radiomics and conventional DL approaches (46).

Treatment response prediction

Chemotherapy is a routine treatment for improving the survival of GC patients. However, the overall efficacy rate of most regimens is less than 40% (84). Recently, conversion therapy, defined as a surgical treatment aimed at R0 resection after NAC for initially unresectable or marginally resectable tumors, emerged as another treatment opinion for metastatic AGC patients (82). Nevertheless, only 20–40% of patients show sensitivity to NAC (85,86). In addition, 40–60% of patients suffered recurrence after R0 surgery (87). Therefore, it is of great importance for identifying optimal candidates who could benefit from AC/NAC and thus avoid unnecessary toxicity. Several teams have investigated the potential of CT-based radiomics and DL methods for predicting AC/NAC response.

Jiang et al. developed a radiomics signature for evaluating tumor infiltrated lymphoid and myeloid cells. Their proposed radiomics signature could also serve as an AC response prediction tool (47). Li et al. built a radiomics signature by extracting features from intratumoral and peritumoral regions of CT images. The radiomics signature showed good predictive ability for DFS and AC response for stage II/III GC (48). Li et al. developed an DL model which consisted of clinical, handcrafted radiomics and DL features for differential diagnosis of signet-ring cell carcinoma (49). The reported AUCs for predicting NAC response for AGC ranged from 0.68 to 0.89 (50,88-92). A significant portion of AGC patients do not achieve pathological downstaging after NAC. Thus, early and precise patient stratification would be beneficial in identifying suitable candidates for NAC. Xu and colleagues proposed a radiomics signature by utilizing restaging CT images for early detection of pathological downstaging with NAC for AGC. The model showed a promising prediction ability with an optimal AUC of 0.96, outperforming the routinely used Response Evaluation Criteria in Solid Tumors (RECIST) system (50).

Liu et al. developed a multi-energy radiomics model by extracting radiomic features from three different sets of monochromatic images derived from DECT. Their findings revealed that high-energy features achieved the best performance for AC response prediction. The predictive ability of the multi-energy model outperformed all the monochromatic radiomics models (51). Similarly, Wang et al. developed a radiomics model based on IU and M images to differentiate serosal invasion after NAC. The model showed great accuracy with AUCs of 0.95 and 0.91 (52).

Recently, analysis of the alteration of radiomics features pre- and post-treatment, refer as delta radiomics, has shown potential for predicting treatment response (93,94). Delta-radiomics is a radiomics-based approach that focuses on the analysis of changes or differences in radiomic features before and after treatment. It aims to capture and quantify the temporal changes in tumor characteristics. By comparing radiomic features extracted from pre- and post-treatment images, delta-radiomics can provide valuable information about treatment response, tumor progression, or the effectiveness of interventions (92). Tan et al. reported a delta radiomics model, revealing moderate predictive accuracy for AC response for AGC patients with an average AUC of 0.75 (53). However, in the study conducted by Chen and colleagues, radiomics features were extracted from 8 image series acquired by DECT, including VP-IU, VP-M, delayed IU, and delayed M images pre- and post-NAC. The pre-NAC IU-based model performed better than the delta radiomics model for predicting DFS and OS for AGC patients who received NAC (95). This result implies that the pretreatment images might provide more information for prognosis.

Immunotherapy, represented by inhibitors of programmed cell death-1 (PD-1), has shown encouraging efficacy in AGC patients. However, effective biomarkers for identifying beneficiary patients are still lacking. We developed and validated a radiomics signature for evaluating tumor-infiltrating regulatory T cells, which contributed to hyperprogression disease after PD-1 treatment (96). Liang et al. recently proposed a radiomics nomogram to predict PD-1 inhibitor response for AGC (54). The model showed good accuracy for identifying responders of anti-PD-1 therapy with AUCs of 0.87 and 0.78. Moreover, the model could also accurately predict the PFS of patients.

Challenges and future perspective

There are still some intractable limitations yet to be addressed in translating AI techniques from bench to bedside.

First and foremost, explainability is the biggest obstacle in translating AI models into routine clinical practice. The explainability in AI refers to the question of whether we can understand the reasoning and decision-making process behind the predictions or decisions made by ML and DL models (97). Due to the complex nature of DL models, which consist of numerous parameters and intricate non-linear functions, their internal workings are often difficult to intuitively interpret. Explainability is crucial for the application of AI, especially in domains where scrutiny, validation, and trust in the model’s predictions are required, such as medical diagnosis (98). Currently, for most ML-based models, especially DL models, it is difficult to explain how the models made certain decisions, which refers to the “black box”. However, transparency and explainability are essential for legal and ethical reasons in clinical medicine. Understanding how an ML-based model comes to its predictions is important to ensure fairness and increase the confidence of clinical practitioners when using it. In addition, AI models can be influenced by data biases, leading to unfair predictions for specific groups or attributes (98). Thus, better explainability can allow developers and end-users to identify potential biases of models and find ways to improve them. Recently, explanatory artificial intelligence (XAI) has drawn great attention from researchers. XAI aims to provide more transparent, explainable decisions for AI models. There are mainly three approaches for improving the explainability of ML models, including proxy models, visualization of CNN, importance estimators (5). Although all these methods have not yet been widely used in the area of medical imaging, further advancement of XAI will possibly help the clinical application of AI models.

Secondly, repeatability and reproducibility are also crucial issues that need to be resolved. Only part of the studies reported the images acquisition, procession, and ML methodology in sufficient detail (15,19,20,22,24-26,29,31,33,36,37,44,47). The scanner parameter and the image pre-processing methods, target area delineation, feature extraction, and model building algorithm varied dramatically among studies. This underlines the need for a standardized methodology for CT acquirement protocol and AI analysis workflow. The Standards for Reporting of Diagnostic Accuracy Studies (STARD) statement recommends structured reporting for studies (99). A team promoted the Image Biomarker Standardization Initiative (IBSI) and provided standardized radiomics features to calibrate different software (100). These initiatives may help to improve the repeatability and reproducibility of research.

Thirdly, most of the current studies were retrospective and carried out in a single-center, which may also lead to the overfitting of the models. Prospective multicenter studies are needed to establish well-curated datasets and ensure the generalizability of the models. However, collecting and harmonizing imaging data from different centers are complex tasks technically. Most of the radiomic features are highly sensitive to the diversity of scanner manufacturers, acquisition protocols, and reconstruction algorithms. Therefore, it is difficult to draw direct comparisons and harmonization of results from different centers. The current standardization of imaging procedures is insufficient to overcome variations in radiomic feature distributions among centers. One popular method is resampling images to a unified voxel size and filtering images to get similar spatial resolution before feature extraction. However, this method may be insufficient to remove the center effect and can be harmful to texture analysis (7,101). DL networks are recently used to synthesize images with similar properties to obtain comparable features. Li et al. proposed a generative adversarial network-based normalization method which could reduce the variability of features obtained by different CT protocols and centers (102). The results showed that radiomic features extracted from harmonized images could effectively improve the accuracy of the LASSO classifier. In another study, by using a public CT texture phantom dataset, DL networks were trained to transform image features to improve their stability across varying CT devices and parameters (103). ComBat is a method for removing batch effects, initially used for genomic analysis. Recently, this method has been used in several multicenter studies for harmonization of radiomic features extracted from different scanners and institutions. The results showed that ComBat could effectively improve the predictive accuracy and stability of models (104-106).

Conclusions

AI-powered tools showed great potential to increase diagnostic accuracy and reduce radiologists’ workload. However, the goal of AI is not to replace human ability but to help oncologists make decisions in their practice. Therefore, radiologists should play a predominant role in AI applications and decide the best ways to integrate these complementary techniques within clinical practice.

Acknowledgments

Funding: This work was supported by the Tianjin Binhai New Area Health Committee Science and Technology Project (No. 2022BWKY017).

Footnote

Reporting Checklist: The authors have completed the Narrative Review reporting checklist. Available at https://tcr.amegroups.com/article/view/10.21037/tcr-23-201/rc

Peer Review File: Available at https://tcr.amegroups.com/article/view/10.21037/tcr-23-201/prf

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tcr.amegroups.com/article/view/10.21037/tcr-23-201/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Sung H, Ferlay J, Siegel RL, et al. Global Cancer Statistics 2020: GLOBOCAN Estimates of Incidence and Mortality Worldwide for 36 Cancers in 185 Countries. CA Cancer J Clin 2021;71:209-49. [Crossref] [PubMed]
2. Rawla P, Barsouk A. Epidemiology of gastric cancer: global trends, risk factors and prevention. Prz Gastroenterol 2019;14:26-38. [Crossref] [PubMed]
3. Tran KA, Kondrashova O, Bradley A, et al. Deep learning in cancer diagnosis, prognosis and treatment selection. Genome Med 2021;13:152. [Crossref] [PubMed]
4. Rajkomar A, Dean J, Kohane I. Machine Learning in Medicine. N Engl J Med 2019;380:1347-58. [Crossref] [PubMed]
5. Reig B, Heacock L, Geras KJ, et al. Machine learning in breast MRI. J Magn Reson Imaging 2020;52:998-1018. [Crossref] [PubMed]
6. Berbís MA, Aneiros-Fernández J, Mendoza Olivares FJ, et al. Role of artificial intelligence in multidisciplinary imaging diagnosis of gastrointestinal diseases. World J Gastroenterol 2021;27:4395-412. [Crossref] [PubMed]
7. Papadimitroulas P, Brocki L, Christopher Chung N, et al. Artificial intelligence: Deep learning in oncological radiomics and challenges of interpretability and data harmonization. Phys Med 2021;83:108-21. [Crossref] [PubMed]
8. LeCun Y, Bengio Y, Hinton G. Deep learning. Nature 2015;521:436-44. [Crossref] [PubMed]
9. Gao Q, Yang L, Lu M, et al. The artificial intelligence and machine learning in lung cancer immunotherapy. J Hematol Oncol 2023;16:55. [Crossref] [PubMed]
10. Suk HI, Lee SW, Shen D, et al. Latent feature representation with stacked auto-encoder for AD/MCI diagnosis. Brain Struct Funct 2015;220:841-59. [Crossref] [PubMed]
11. Kido S, Hirano Y, Mabu S. Deep Learning for Pulmonary Image Analysis: Classification, Detection, and Segmentation. Adv Exp Med Biol 2020;1213:47-58. [Crossref] [PubMed]
12. Wang S, Shi J, Ye Z, et al. Predicting EGFR mutation status in lung adenocarcinoma on computed tomography image using deep learning. Eur Respir J 2019;53:1800986. [Crossref] [PubMed]
13. Hsiao YJ, Wen YC, Lai WY, et al. Application of artificial intelligence-driven endoscopic screening and diagnosis of gastric cancer. World J Gastroenterol 2021;27:2979-93. [Crossref] [PubMed]
14. Bi WL, Hosny A, Schabath MB, et al. Artificial intelligence in cancer imaging: Clinical challenges and applications. CA Cancer J Clin 2019;69:127-57. [Crossref] [PubMed]
15. Gao X, Ma T, Cui J, et al. A radiomics-based model for prediction of lymph node metastasis in gastric cancer. Eur J Radiol 2020;129:109069. [Crossref] [PubMed]
16. Hu H, Gong L, Dong D, et al. Identifying early gastric cancer under magnifying narrow-band images with deep learning: a multicenter study. Gastrointest Endosc 2021;93:1333-41.e3. [Crossref] [PubMed]
17. Gao X, Ma T, Cui J, et al. A CT-based Radiomics Model for Prediction of Lymph Node Metastasis in Early Stage Gastric Cancer. Acad Radiol 2021;28:e155-64. [Crossref] [PubMed]
18. Chen X, Yang Z, Yang J, et al. Radiomics analysis of contrast-enhanced CT predicts lymphovascular invasion and disease outcome in gastric cancer: a preliminary study. Cancer Imaging 2020;20:24. [Crossref] [PubMed]
19. Wang Y, Liu W, Yu Y, et al. Prediction of the Depth of Tumor Invasion in Gastric Cancer: Potential Role of CT Radiomics. Acad Radiol 2020;27:1077-84. [Crossref] [PubMed]
20. Sun RJ, Fang MJ, Tang L, et al. CT-based deep learning radiomics analysis for evaluation of serosa invasion in advanced gastric cancer. Eur J Radiol 2020;132:109277. [Crossref] [PubMed]
21. Ma Z, Fang M, Huang Y, et al. CT-based radiomics signature for differentiating Borrmann type IV gastric cancer from primary gastric lymphoma. Eur J Radiol 2017;91:142-7. [Crossref] [PubMed]
22. Feng B, Huang L, Liu Y, et al. A Transfer Learning Radiomics Nomogram for Preoperative Prediction of Borrmann Type IV Gastric Cancer From Primary Gastric Lymphoma. Front Oncol 2022;11:802205. [Crossref] [PubMed]
23. Wang R, Liu H, Liang P, et al. Radiomics analysis of CT imaging for differentiating gastric neuroendocrine carcinomas from gastric adenocarcinomas. Eur J Radiol 2021;138:109662. [Crossref] [PubMed]
24. Chen T, Wu J, Cui C, et al. CT-based radiomics nomograms for preoperative prediction of diffuse-type and signet ring cell gastric cancer: a multicenter development and validation cohort. J Transl Med 2022;20:38. [Crossref] [PubMed]
25. Dong D, Fang MJ, Tang L, et al. Deep learning radiomic nomogram can predict the number of lymph node metastasis in locally advanced gastric cancer: an international multicenter study. Ann Oncol 2020;31:912-20. [Crossref] [PubMed]
26. Wang Y, Liu W, Yu Y, et al. CT radiomics nomogram for the preoperative prediction of lymph node metastasis in gastric cancer. Eur Radiol 2020;30:976-86. [Crossref] [PubMed]
27. Li J, Dong D, Fang M, et al. Dual-energy CT-based deep learning radiomics can improve lymph node metastasis risk prediction for gastric cancer. Eur Radiol 2020;30:2324-33. [Crossref] [PubMed]
28. Jin C, Jiang Y, Yu H, et al. Deep learning analysis of the primary tumour and the prediction of lymph node metastases in gastric cancer. Br J Surg 2021;108:542-9. [Crossref] [PubMed]
29. Fan L, Li J, Zhang H, et al. Machine learning analysis for the noninvasive prediction of lymphovascular invasion in gastric cancer using PET/CT and enhanced CT-based radiomics and clinical variables. Abdom Radiol (NY) 2022;47:1209-22. [Crossref] [PubMed]
30. Liu S, He J, Liu S, et al. Radiomics analysis using contrast-enhanced CT for preoperative prediction of occult peritoneal metastasis in advanced gastric cancer. Eur Radiol 2020;30:239-46. [Crossref] [PubMed]
31. Dong D, Tang L, Li ZY, et al. Development and validation of an individualized nomogram to identify occult peritoneal metastasis in patients with advanced gastric cancer. Ann Oncol 2019;30:431-8. [Crossref] [PubMed]
32. Huang W, Zhou K, Jiang Y, et al. Radiomics Nomogram for Prediction of Peritoneal Metastasis in Patients With Gastric Cancer. Front Oncol 2020;10:1416. [Crossref] [PubMed]
33. Mirniaharikandehei S, Heidari M, Danala G, et al. Applying a random projection algorithm to optimize machine learning model for predicting peritoneal metastasis in gastric cancer patients using CT images. Comput Methods Programs Biomed 2021;200:105937. [Crossref] [PubMed]
34. Chen Y, Xi W, Yao W, et al. Dual-Energy Computed Tomography-Based Radiomics to Predict Peritoneal Metastasis in Gastric Cancer. Front Oncol 2021;11:659981. [Crossref] [PubMed]
35. Liu D, Zhang W, Hu F, et al. A Bounding Box-Based Radiomics Model for Detecting Occult Peritoneal Metastasis in Advanced Gastric Cancer: A Multicenter Study. Front Oncol 2021;11:777760. [Crossref] [PubMed]
36. Huang Z, Liu D, Chen X, et al. Deep Convolutional Neural Network Based on Computed Tomography Images for the Preoperative Diagnosis of Occult Peritoneal Metastasis in Advanced Gastric Cancer. Front Oncol 2020;10:601869. [Crossref] [PubMed]
37. Jiang Y, Liang X, Wang W, et al. Noninvasive Prediction of Occult Peritoneal Metastasis in Gastric Cancer Using Deep Learning. JAMA Netw Open 2021;4:e2032269. [Crossref] [PubMed]
38. Zhao H, Li W, Lyu P, et al. TCGA-TCIA-Based CT Radiomics Study for Noninvasively Predicting Epstein-Barr Virus Status in Gastric Cancer. AJR Am J Roentgenol 2021;217:124-34. [Crossref] [PubMed]
39. Zhang C, Wen HL, Zhang R, et al. Computed tomography radiomics to predict EBER positivity in Epstein-Barr virus-associated gastric adenocarcinomas: a retrospective study. Acta Radiol 2022;63:1005-13. [Crossref] [PubMed]
40. Wang Y, Yu Y, Han W, et al. CT Radiomics for Distinction of Human Epidermal Growth Factor Receptor 2 Negative Gastric Cancer. Acad Radiol 2021;28:e86-92. [Crossref] [PubMed]
41. Li W, Zhang L, Tian C, et al. Prognostic value of computed tomography radiomics features in patients with gastric cancer following curative resection. Eur Radiol 2019;29:3079-89. [Crossref] [PubMed]
42. Jiang Y, Chen C, Xie J, et al. Radiomics signature of computed tomography imaging for prediction of survival and chemotherapeutic benefits in gastric cancer. EBioMedicine 2018;36:171-82. [Crossref] [PubMed]
43. Jin Y, Xu Y, Li Y, et al. Integrative Radiogenomics Approach for Risk Assessment of Postoperative and Adjuvant Chemotherapy Benefits for Gastric Cancer Patients. Front Oncol 2021;11:755271. [Crossref] [PubMed]
44. Shin J, Lim JS, Huh YM, et al. A radiomics-based model for predicting prognosis of locally advanced gastric cancer in the preoperative setting. Sci Rep 2021;11:1879. [Crossref] [PubMed]
45. Jiang Y, Jin C, Yu H, et al. Development and Validation of a Deep Learning CT Signature to Predict Survival and Chemotherapy Benefit in Gastric Cancer: A Multicenter, Retrospective Study. Ann Surg 2021;274:e1153-61. [Crossref] [PubMed]
46. Zhang L, Dong D, Zhong L, et al. Multi-Focus Network to Decode Imaging Phenotype for Overall Survival Prediction of Gastric Cancer Patients. IEEE J Biomed Health Inform 2021;25:3933-42. [Crossref] [PubMed]
47. Jiang Y, Wang H, Wu J, et al. Noninvasive imaging evaluation of tumor immune microenvironment to predict outcomes in gastric cancer. Ann Oncol 2020;31:760-8. [Crossref] [PubMed]
48. Li J, Zhang C, Wei J, et al. Intratumoral and Peritumoral Radiomics of Contrast-Enhanced CT for Prediction of Disease-Free Survival and Chemotherapy Response in Stage II/III Gastric Cancer. Front Oncol 2020;10:552270. [Crossref] [PubMed]
49. Li C, Qin Y, Zhang WH, et al. Deep learning-based AI model for signet-ring cell carcinoma diagnosis and chemotherapy response prediction in gastric cancer. Med Phys 2022;49:1535-46. [Crossref] [PubMed]
50. Xu Q, Sun Z, Li X, et al. Advanced gastric cancer: CT radiomics prediction and early detection of downstaging with neoadjuvant chemotherapy. Eur Radiol 2021;31:8765-74. [Crossref] [PubMed]
51. Liu YY, Zhang H, Wang L, et al. Predicting Response to Systemic Chemotherapy for Advanced Gastric Cancer Using Pre-Treatment Dual-Energy CT Radiomics: A Pilot Study. Front Oncol 2021;11:740732. [Crossref] [PubMed]
52. Wang L, Zhang Y, Chen Y, et al. The Performance of a Dual-Energy CT Derived Radiomics Model in Differentiating Serosal Invasion for Advanced Gastric Cancer Patients After Neoadjuvant Chemotherapy: Iodine Map Combined With 120-kV Equivalent Mixed Images. Front Oncol 2021;10:562945. [Crossref] [PubMed]
53. Tan JW, Wang L, Chen Y, et al. Predicting Chemotherapeutic Response for Far-advanced Gastric Cancer by Radiomics with Deep Learning Semi-automatic Segmentation. J Cancer 2020;11:7224-36. [Crossref] [PubMed]
54. Liang Z, Huang A, Wang L, et al. A radiomics model predicts the response of patients with advanced gastric cancer to PD-1 inhibitor treatment. Aging (Albany NY) 2022;14:907-22. [Crossref] [PubMed]
55. Wang FH, Shen L, Li J, et al. The Chinese Society of Clinical Oncology (CSCO): clinical guidelines for the diagnosis and treatment of gastric cancer. Cancer Commun (Lond) 2019;39:10. [Crossref] [PubMed]
56. Fukagawa T, Katai H, Mizusawa J, et al. A prospective multi-institutional validity study to evaluate the accuracy of clinical diagnosis of pathological stage III gastric cancer (JCOG1302A). Gastric Cancer 2018;21:68-73. [Crossref] [PubMed]
57. Kelessis NG, Vassilopoulos PP, Bai MP, et al. Update of the role of surgery in the multimodal treatment of MALT gastric lymphomas. Anticancer Res 2002;22:3457-63. [PubMed]
58. Kim HE, Cosa-Linan A, Santhanam N, et al. Transfer learning for medical image classification: a literature review. BMC Med Imaging 2022;22:69. [Crossref] [PubMed]
59. Xue LY, Jiang ZY, Fu TT, et al. Transfer learning radiomics based on multimodal ultrasound imaging for staging liver fibrosis. Eur Radiol 2020;30:2973-83. [Crossref] [PubMed]
60. Lo SS, Wu CW, Chen JH, et al. Surgical results of early gastric cancer and proposing a treatment strategy. Ann Surg Oncol 2007;14:340-7. [Crossref] [PubMed]
61. Kim AY, Kim HJ, Ha HK. Gastric cancer by multidetector row CT: preoperative staging. Abdom Imaging 2005;30:465-72. [Crossref] [PubMed]
62. Huang YQ, Liang CH, He L, et al. Development and Validation of a Radiomics Nomogram for Preoperative Prediction of Lymph Node Metastasis in Colorectal Cancer. J Clin Oncol 2016;34:2157-64. [Crossref] [PubMed]
63. Feng QX, Liu C, Qi L, et al. An Intelligent Clinical Decision Support System for Preoperative Prediction of Lymph Node Metastasis in Gastric Cancer. J Am Coll Radiol 2019;16:952-60. [Crossref] [PubMed]
64. Gao Y, Zhang ZD, Li S, et al. Deep neural network-assisted computed tomography diagnosis of metastatic lymph nodes from gastric cancer. Chin Med J (Engl) 2019;132:2804-11. [Crossref] [PubMed]
65. Jiang Y, Wang W, Chen C, et al. Radiomics Signature on Computed Tomography Imaging: Association With Lymph Node Metastasis in Patients With Gastric Cancer. Front Oncol 2019;9:340. [Crossref] [PubMed]
66. Sun Z, Jiang Y, Chen C, et al. Radiomics signature based on computed tomography images for the preoperative prediction of lymph node metastasis at individual stations in gastric cancer: A multicenter study. Radiother Oncol 2021;165:179-90. [Crossref] [PubMed]
67. Zhang CD, Ning FL, Zeng XT, et al. Lymphovascular invasion as a predictor for lymph node metastasis and a prognostic factor in gastric cancer patients under 70 years of age: A retrospective analysis. Int J Surg 2018;53:214-20. [Crossref] [PubMed]
68. Wada T, Kunisaki C, Hasegawa S, et al. Factors predictive of recurrence after surgery for gastric cancer followed by adjuvant S-1 chemotherapy. Anticancer Res 2013;33:1747-51. [PubMed]
69. Li Q, Feng QX, Qi L, et al. Prognostic aspects of lymphovascular invasion in localized gastric cancer: new insights into the radiomics and deep transfer learning from contrast-enhanced CT imaging. Abdom Radiol (NY) 2022;47:496-507. [Crossref] [PubMed]
70. Coccolini F, Cotte E, Glehen O, et al. Intraperitoneal chemotherapy in advanced gastric cancer. Meta-analysis of randomized trials. Eur J Surg Oncol 2014;40:12-26. [Crossref] [PubMed]
71. Kim SJ, Kim HH, Kim YH, et al. Peritoneal metastasis: detection with 16- or 64-detector row CT in patients undergoing surgery for gastric cancer. Radiology 2009;253:407-15. [Crossref] [PubMed]
72. Wang S, Wang Q, Shao Y, et al. Iterative Label Denoising Network: Segmenting Male Pelvic Organs in CT From 3D Bounding Box Annotations. IEEE Trans Biomed Eng 2020;67:2710-20. [Crossref] [PubMed]
73. Saha BN, Ray N, Greiner R, et al. Quick detection of brain tumors and edemas: a bounding box method using symmetry. Comput Med Imaging Graph 2012;36:95-107. [Crossref] [PubMed]
74. Comprehensive molecular characterization of gastric adenocarcinoma. Nature 2014;513:202-9. [Crossref] [PubMed]
75. Washington MK, Goldberg RM, Chang GJ, et al. Diagnosis of digestive system tumours. Int J Cancer 2021;148:1040-50. [Crossref] [PubMed]
76. Cristescu R, Lee J, Nebozhyn M, et al. Molecular analysis of gastric cancer identifies subtypes associated with distinct clinical outcomes. Nat Med 2015;21:449-56. [Crossref] [PubMed]
77. Sohn BH, Hwang JE, Jang HJ, et al. Clinical Significance of Four Molecular Subtypes of Gastric Cancer Identified by The Cancer Genome Atlas Project. Clin Cancer Res 2017;23:4441-9. [Crossref] [PubMed]
78. Kim HS, Shin SJ, Beom SH, et al. Comprehensive expression profiles of gastric cancer molecular subtypes by immunohistochemistry: implications for individualized therapy. Oncotarget 2016;7:44608-20. [Crossref] [PubMed]
79. Wang Y, He L, Cheng Y. An independent survival prognostic role for human epidermal growth factor receptor 2 in gastric cancer: evidence from a meta-analysis. Clin Transl Oncol 2018;20:212-20. [Crossref] [PubMed]
80. Luo X, He Y, Tang H, et al. Effects of HER2 on the invasion and migration of gastric cancer. Am J Transl Res 2019;11:7604-13. [PubMed]
81. Bang YJ, Van Cutsem E, Feyereislova A, et al. Trastuzumab in combination with chemotherapy versus chemotherapy alone for treatment of HER2-positive advanced gastric or gastro-oesophageal junction cancer (ToGA): a phase 3, open-label, randomised controlled trial. Lancet 2010;376:687-97. [Crossref] [PubMed]
82. Kinoshita J, Yamaguchi T, Moriyama H, et al. Current status of conversion surgery for stage IV gastric cancer. Surg Today 2021;51:1736-54. [Crossref] [PubMed]
83. Wakatsuki T, Yamamoto N, Sano T, et al. Clinical impact of intratumoral HER2 heterogeneity on trastuzumab efficacy in patients with HER2-positive gastric cancer. J Gastroenterol 2018;53:1186-95. [Crossref] [PubMed]
84. Van Cutsem E, Sagaert X, Topal B, et al. Gastric cancer. Lancet 2016;388:2654-64. [Crossref] [PubMed]
85. Al-Batran SE, Hofheinz RD, Pauligk C, et al. Histopathological regression after neoadjuvant docetaxel, oxaliplatin, fluorouracil, and leucovorin versus epirubicin, cisplatin, and fluorouracil or capecitabine in patients with resectable gastric or gastro-oesophageal junction adenocarcinoma (FLOT4-AIO): results from the phase 2 part of a multicentre, open-label, randomised phase 2/3 trial. Lancet Oncol 2016;17:1697-708. [Crossref] [PubMed]
86. Al-Batran SE, Homann N, Pauligk C, et al. Perioperative chemotherapy with fluorouracil plus leucovorin, oxaliplatin, and docetaxel versus fluorouracil or capecitabine plus cisplatin and epirubicin for locally advanced, resectable gastric or gastro-oesophageal junction adenocarcinoma (FLOT4): a randomised, phase 2/3 trial. Lancet 2019;393:1948-57. [Crossref] [PubMed]
87. Wagner AD, Syn NL, Moehler M, et al. Chemotherapy for advanced gastric cancer. Cochrane Database Syst Rev 2017;8:CD004064. [PubMed]
88. Sun KY, Hu HT, Chen SL, et al. CT-based radiomics scores predict response to neoadjuvant chemotherapy and survival in patients with gastric cancer. BMC Cancer 2020;20:468. [Crossref] [PubMed]
89. Xie K, Cui Y, Zhang D, et al. Pretreatment Contrast-Enhanced Computed Tomography Radiomics for Prediction of Pathological Regression Following Neoadjuvant Chemotherapy in Locally Advanced Gastric Cancer: A Preliminary Multicenter Study. Front Oncol 2022;11:770758. [Crossref] [PubMed]
90. Wang W, Peng Y, Feng X, et al. Development and Validation of a Computed Tomography-Based Radiomics Signature to Predict Response to Neoadjuvant Chemotherapy for Locally Advanced Gastric Cancer. JAMA Netw Open 2021;4:e2121143. [Crossref] [PubMed]
91. Chen Y, Wei K, Liu D, et al. A Machine Learning Model for Predicting a Major Response to Neoadjuvant Chemotherapy in Advanced Gastric Cancer. Front Oncol 2021;11:675458. [Crossref] [PubMed]
92. Li Z, Zhang D, Dai Y, et al. Computed tomography-based radiomics for prediction of neoadjuvant chemotherapy outcomes in locally advanced gastric cancer: A pilot study. Chin J Cancer Res 2018;30:406-14. [Crossref] [PubMed]
93. Rao SX, Lambregts DM, Schnerr RS, et al. CT texture analysis in colorectal liver metastases: A better way than size and volume measurements to assess response to chemotherapy? United European Gastroenterol J 2016;4:257-63. [Crossref] [PubMed]
94. Fave X, Zhang L, Yang J, et al. Delta-radiomics features for the prediction of patient outcomes in non-small cell lung cancer. Sci Rep 2017;7:588. [Crossref] [PubMed]
95. Chen Y, Yuan F, Wang L, et al. Evaluation of dual-energy CT derived radiomics signatures in predicting outcomes in patients with advanced gastric cancer after neoadjuvant chemotherapy. Eur J Surg Oncol 2022;48:339-47. [Crossref] [PubMed]
96. Gao X, Ma T, Bai S, et al. A CT-based radiomics signature for evaluating tumor infiltrating Treg cells and outcome prediction of gastric cancer. Ann Transl Med 2020;8:469. [Crossref] [PubMed]
97. Linardatos P, Papastefanopoulos V, Kotsiantis S, Explainable AI. A Review of Machine Learning Interpretability Methods. Entropy (Basel) 2020;23:18. [Crossref] [PubMed]
98. Zhang Y, Weng Y, Lund J. Applications of Explainable Artificial Intelligence in Diagnosis and Surgery. Diagnostics (Basel) 2022;12:237. [Crossref] [PubMed]
99. Bossuyt PM, Reitsma JB, Bruns DE, et al. STARD 2015: An Updated List of Essential Items for Reporting Diagnostic Accuracy Studies. Radiology 2015;277:826-32. [Crossref] [PubMed]
100. Zwanenburg A, Vallières M, Abdalah MA, et al. The Image Biomarker Standardization Initiative: Standardized Quantitative Radiomics for High-Throughput Image-based Phenotyping. Radiology 2020;295:328-38. [Crossref] [PubMed]
101. Orlhac F, Eertink JJ, Cottereau AS, et al. A Guide to ComBat Harmonization of Imaging Biomarkers in Multicenter Studies. J Nucl Med 2022;63:172-9. [Crossref] [PubMed]
102. Li Y, Han G, Wu X, et al. Normalization of multicenter CT radiomics by a generative adversarial network method. Phys Med Biol 2021;66: [Crossref] [PubMed]
103. Andrearczyk V, Depeursinge A, Müller H. Neural network training for cross-protocol radiomic feature standardization in computed tomography. J Med Imaging (Bellingham) 2019;6:024008. [Crossref] [PubMed]
104. Dissaux G, Visvikis D, Da-Ano R, et al. Pretreatment (18)F-FDG PET/CT Radiomics Predict Local Recurrence in Patients Treated with Stereotactic Body Radiotherapy for Early-Stage Non-Small Cell Lung Cancer: A Multicentric Study. J Nucl Med 2020;61:814-20. [Crossref] [PubMed]
105. Wu Q, Wang S, Li L, et al. Radiomics Analysis of Computed Tomography helps predict poor prognostic outcome in COVID-19. Theranostics 2020;10:7231-44. [Crossref] [PubMed]
106. Garau N, Paganelli C, Summers P, et al. External validation of radiomics-based predictive models in low-dose CT screening for early lung cancer diagnosis. Med Phys 2020;47:4125-36. [Crossref] [PubMed]