Recent application of artificial intelligence on histopathologic image-based prediction of gene mutation in solid cancers.

PURPOSE: Evaluation of genetic mutations in cancers is important because distinct mutational profiles help determine individualized drug therapy. However, molecular analyses are not routinely performed in all cancers because they are expensive, time-consuming and not universally available. Artificial intelligence (AI) has shown the potential to determine a wide range of genetic mutations on histologic image analysis. Here, we assessed the status of mutation prediction AI models on histologic images by a systematic review. METHODS: A literature search using the MEDLINE, Embase and Cochrane databases was conducted in August 2021. The articles were shortlisted by titles and abstracts. After a full-text review, publication trends, study characteristic analysis and comparison of performance metrics were performed. RESULTS: Twenty-four studies were found mostly from developed countries, and their number is increasing. The major targets were gastrointestinal, genitourinary, gynecological, lung and head and neck cancers. Most studies used the Cancer Genome Atlas, with a few using an in-house dataset. The area under the curve of some of the cancer driver gene mutations in particular organs was satisfactory, such as 0.92 of BRAF in thyroid cancers and 0.79 of EGFR in lung cancers, whereas the average of all gene mutations was 0.64, which is still suboptimal. CONCLUSION: AI has the potential to predict gene mutations on histologic images with appropriate caution. Further validation with larger datasets is still required before AI models can be used in clinical practice to predict gene mutations.

Assessing risk stratification in long-term outcomes of rectal neuroendocrine tumors following endoscopic resection: a multicenter retrospective study.

OBJECTIVES: While endoscopic resection of rectal neuroendocrine tumors (NETs) has significantly increased, long-term data on risk factors for recurrence are still lacking. Our aim is to analyze the long-term outcomes of patients with rectal NETs after endoscopic resection through risk stratification. METHODS: In this multicenter retrospective study, we included patients who underwent endoscopic resection of rectal NETs from 2009 to 2018 and were followed for ≥12 months at five university hospitals. We classified the patients into three risk groups according to the clinicopathological status of the rectal neuroendocrine tumors: low, indeterminate, and high. The high-risk group was defined if the tumors have any of the followings: size ≥ 10 mm, lymphovascular invasion, muscularis propria or deeper invasion, positive resection margins, or mitotic count ≥2/10. RESULTS: A total of 346 patients were included, with 144 (41.6%), 121 (35.0%), and 81 (23.4%) classified into the low-, indeterminate-, and high-risk groups, respectively. Among the high-risk group, seven patients (8.6%) received salvage treatment 28 (27-67) days after the initial endoscopic resection, with no reported extracolonic recurrence. Throughout the follow-up period, 1.1% (4/346) of patients experienced extracolonic recurrences at 56.5 (54-73) months after the initial endoscopic resection. Three of these patients (75%) were in the high-risk group and did not undergo salvage treatment. The risk of extracolonic recurrence was significantly higher in the high-risk group compared to the other groups (p = 0.039). CONCLUSION: Physicians should be concerned about the possibility of metastasis during long-term follow-up of high-risk patients and consider salvage treatment.

Deep learning captures selective features for discrimination of microsatellite instability from pathologic tissue slides of gastric cancer.

Microsatellite instability (MSI) status is an important prognostic marker for various cancers. Furthermore, because immune checkpoint inhibitors are much more effective in tumors with high level of MSI (MSI-H), MSI status is routinely tested in multiple cancer types. Therefore, many studies have tested the feasibility of deep learning (DL)-based prediction of MSI status from hematoxylin and eosin (H&E)-stained tissue slides. In the present study, we attempted a fully automated classification of MSI status in gastric cancer (GC) tissue slides. For frozen and formalin-fixed paraffin-embedded (FFPE) GC tissues from The Cancer Genome Atlas (TCGA), the areas under the curves (AUCs) for the receiver operating characteristic (ROC) curves were 0.893 and 0.902, respectively. The classifier trained with the TCGA FFPE tissues performed well on an external validation Asian FFPE cohort, with an AUC of 0.874. However, the DL-based classifier seems incompatible with cancers from different organs because morphologic features of MSI-H tissues are different. Analysis of histomorphologic features of MSI-H GC tissues suggested that MSI-H GC could largely be divided into two groups: intestinal type tumors with moderate to poor differentiation and diffuse type mucinous tumors. However, the recognizable morphologic features cannot completely explain the good performance of the DL-based classifier. These results indicate that DL could automatically learn the optimal features for discrimination of MSI status in GC tissue slides. This study demonstrated the potential of a DL-based MSI classifier as a screening tool for definitive cases.

Anchoring endoscopic mucosal resection versus conventional endoscopic mucosal resection for large nonpedunculated colorectal polyps: a randomized controlled trial.

BACKGROUND : Colorectal polyps > 10 mm in size are often incompletely resected. Anchoring-endoscopic mucosal resection (A-EMR) is the technique of making a small incision at the oral side of the polyp using a snare tip after submucosal injection to avoid slippage during ensnaring. This study was performed to evaluate whether A-EMR could increase the complete resection rate for large colorectal polyps compared with conventional endoscopic mucosal resection (C-EMR). METHODS : Polyps with sizes of 10-25 mm were randomly allocated to either the A-EMR or the C-EMR groups. RESULTS : 105 and 106 polyps were resected using A-EMR and C-EMR, respectively. In the intention-to-treat population, the complete resection rate was 89.5 % in the A-EMR group and 74.5 % in the C-EMR group (relative risk [RR] 1.20, 95 %CI 1.04 to 1.38; P = 0.01). The en bloc resection rates for the A-EMR and C-EMR groups were 92.4 % vs. 76.4 % (RR 1.21, 95 %CI 1.06 to 1.37; P = 0.005) and R0 resection rates were 77.1 % vs. 64.2 % (RR 1.18, 95 %CI 0.98 to 1.42; P = 0.07), respectively. The median (interquartile range [IQR]) total procedure time was 3.2 (2.6-4.1) minutes in the A-EMR group and 3.0 (2.2-4.6) minutes in the C-EMR group (median difference 0.2 minutes, 95 %CI -0.22 to 0.73; P = 0.25). There was one episode of delayed bleeding and one perforation in the C-EMR group. CONCLUSIONS : A-EMR was superior to C-EMR for the complete resection of large colorectal polyps. A-EMR can be considered one of the standard methods for the removal of colorectal polyps of 10 mm or more in size.

Long-term outcomes after endoscopic versus surgical resection of T1 colorectal carcinoma.

BACKGROUND: The long-term outcomes of patients with T1 colorectal cancer (CRC) who undergo endoscopic and/or surgical treatment are not well understood. Invasive CRC confined to the colonic submucosa (T1 CRC) is challenging in terms of clinical decision-making. We compared the long-term outcomes of T1 CRC by treatment method. METHODS: We examined 370 patients with pathological T1 CRC treated between 2000 and 2015 at Seoul St. Mary's Hospital. In total, 93 patients underwent endoscopic resection (ER) only, 82 underwent additional surgery after ER, and 175 underwent surgical resection only. Patients who did not meet the curative criteria were defined as "high-risk." High-risk patients were classified into three groups according to the treatment modalities: ER only (Group A: 35 patients), additional surgery after ER (Group B: 72 patients), and surgical resection only (Group C: 133 patients). The recurrence-free and overall survival (OS) rates, and factors associated with recurrence and mortality, were analyzed. Factors associated with lymph node metastasis (LNM) were subjected to multivariate analysis. RESULTS: Of the 370 patients, 7 experienced recurrence and 7 died. All recurrences occurred in the high-risk group and two deaths were in the low-risk group. In high-risk groups, there was no significant group difference in recurrence-free survival (P = 0.511) or OS (P =0.657). Poor histology (P =0.042) was associated with recurrence, and vascular invasion (P =0.044) with mortality. LNMs were observed in 30 of 277 patients who underwent surgery either initially or secondarily. Lymphatic invasion was significantly associated with the incidence of LNM (P < 0.001). CONCLUSIONS: ER prior to surgery did not affect the prognosis of high-risk T1 CRC patients, and did not worsen the clinical outcomes of patients who required additional surgery. Lymphatic invasion was the most important predictor of LNM.

Granulomatous gastritis induced by chronic peanut ingestion.

Granulomatous gastritis (GG) is a rare condition, with incidence between 0.08 and 0.35% in gastric biopsies. Various infectious and non-infectious aetiologies can be considered to cause granulomatous gastritis. Foreign bodies are a rare aetiology of GG and may result from foods, suture materials, or medications. We report a 59-year-old woman who had eaten large amounts of peanuts for more than 10 years and presented with epigastric discomfort. Esophagogastroduodenoscopy revealed multiple nodular lesions with ulcer scars at the stomach, which was diagnosed as GG probably caused by chronic peanut ingestion on endoscopic mucosal resection.

Rainwater-Removal Image Conversion Learning with Training Pair Augmentation.

In this study, we proposed an image conversion method that efficiently removes raindrops on a camera lens from an image using a deep learning technique. The proposed method effectively presents a raindrop-removed image using the Pix2pix generative adversarial network (GAN) model, which can understand the characteristics of two images in terms of newly formed images of different domains. The learning method based on the captured image has the disadvantage that a large amount of data is required for learning and that unnecessary noise is generated owing to the nature of the learning model. In particular, obtaining sufficient original and raindrops images is the most important aspect of learning. Therefore, we proposed a method that efficiently obtains learning data by generating virtual water-drop image data and effectively identifying it using a convolutional neural network (CNN).

Combined high NEDD9 expression and E-cadherin loss correlate with poor clinical outcome in gastric cancer.

BACKGROUND AND AIM: Neural precursor cell expressed developmentally downregulated 9 (NEDD9) is a member of the Cas family. Previous studies have revealed that NEDD9 coordinates the focal adhesion kinase and Src signaling cascades that are involved in integrin-dependent adhesion and migration, invasion, cell apoptosis and life cycle, and survival, which may play a role in epithelial-mesenchymal transformation. The aim of this study was to analyze the expression of NEDD9 and E-cadherin in gastric cancer (GC) and evaluate their clinical significance. METHODS: NEDD9 and E-cadherin expression was analyzed with immunohistochemistry using tissue microarray technique in 435 GC patients who underwent gastrectomy. The NEDD9 expression level was defined by the combination score, which was determined by multiplying the staining intensity score and the proportion score (≥5; NEDD9-high, <5; NEDD9-low). E-cadherin loss was defined as a total loss of staining. The clinicopathologic parameters, overall survival, and disease-free survival rates were analyzed according to the NEDD9 and E-cadherin expression status. RESULTS: The combined NEDD9 and E-cadherin expression status correlated with lymphatic invasion (P = 0.001), vascular invasion (P = 0.020), and T stage (P = 0.001). Combined high NEDD9 expression and loss of E-cadherin expression status had a worse overall survival rate (P < 0.001) and served as a poor prognostic factor (Hazard ratio 2.49, 95% CI 1.25-5, P = 0.01). CONCLUSIONS: Immunohistochemical staining for NEDD9 and E-cadherin may function as a candidate prognostic marker for gastric cancer in everyday practice, especially when applied in combination.

Circumferential submucosal incision prior to endoscopic mucosal resection versus conventional endoscopic mucosal resection for colorectal lesions with endoscopic features of sessile serrated lesions.

BACKGROUND AND AIMS: Sessile serrated lesions (SSLs) are more prone to incomplete resection than conventional adenomas. This study evaluated whether circumferential submucosal incision prior to endoscopic mucosal resection (CSI-EMR) can increase the rate of complete and en bloc resections of colorectal lesions with endoscopic features of SSL. METHODS: Retrospective analyses and propensity score matching were performed for the resection of colorectal lesions ≥ 10 mm with endoscopic features of SSL. RESULTS: After 1:1 ratio matching, 127 lesions in the CSI-EMR group and 127 in the EMR group were selected for analysis. The median size of the lesions was 15 mm (IQR 12-16) in both groups. There was no significant difference in either the complete resection rate or en bloc resection rate between CSI-EMR and EMR groups (96.9% vs. 92.9%, P = 0.155; 92.1% vs. 89.0%, P = 0.391). By contrast, the R0 resection rate was significantly higher in the CSI-EMR group than in the EMR group (89.8% vs. 59.8%, P < 0.001). The median procedure time was significantly longer in the CSI-EMR group than in the EMR group (6.28 min vs. 2.55 min, P < 0.001), whereas there was no significant difference between the two groups in the incidence of adverse events or recurrence rate. Multivariate analysis showed that CSI-EMR was the only factor significantly associated with R0 resection (P < 0.001). CONCLUSIONS: For colorectal lesions with endoscopic features of SSL, CSI-EMR does not increase the complete or en bloc resection rate, but does increase the R0 resection rate. The procedure time is longer for CSI-EMR than EMR. The association of CSI-EMR with R0 resection and non-recurrence should be further evaluated.

Tone Image Classification and Weighted Learning for Visible and NIR Image Fusion.

In this paper, to improve the slow processing speed of the rule-based visible and NIR (near-infrared) image synthesis method, we present a fast image fusion method using DenseFuse, one of the CNN (convolutional neural network)-based image synthesis methods. The proposed method applies a raster scan algorithm to secure visible and NIR datasets for effective learning and presents a dataset classification method using luminance and variance. Additionally, in this paper, a method for synthesizing a feature map in a fusion layer is presented and compared with the method for synthesizing a feature map in other fusion layers. The proposed method learns the superior image quality of the rule-based image synthesis method and shows a clear synthesized image with better visibility than other existing learning-based image synthesis methods. Compared with the rule-based image synthesis method used as the target image, the proposed method has an advantage in processing speed by reducing the processing time to three times or more.

Feasibility of deep learning-based fully automated classification of microsatellite instability in tissue slides of colorectal cancer.

High levels of microsatellite instability (MSI-H) occurs in about 15% of sporadic colorectal cancer (CRC) and is an important predictive marker for response to immune checkpoint inhibitors. To test the feasibility of a deep learning (DL)-based classifier as a screening tool for MSI status, we built a fully automated DL-based MSI classifier using pathology whole-slide images (WSIs) of CRCs. On small image patches of The Cancer Genome Atlas (TCGA) CRC WSI dataset, tissue/non-tissue, normal/tumor and MSS/MSI-H classifiers were applied sequentially for the fully automated prediction of the MSI status. The classifiers were also tested on an independent cohort. Furthermore, to test how the expansion of the training data affects the performance of the DL-based classifier, additional classifier trained on both TCGA and external datasets was tested. The areas under the receiver operating characteristic curves were 0.892 and 0.972 for the TCGA and external datasets, respectively, by a classifier trained on both datasets. The performance of the DL-based classifier was much better than that of previously reported histomorphology-based methods. We speculated that about 40% of CRC slides could be screened for MSI status without molecular testing by the DL-based classifier. These results demonstrated that the DL-based method has potential as a screening tool to discriminate molecular alteration in tissue slides.

The development and validation of a predictive model for recurrence in rectal cancer based on radiological and clinicopathological data.

OBJECTIVE: To develop a prediction model for recurrence by incorporating radiological and clinicopathological prognostic factors in rectal cancer patients. METHODS: All radiologic and clinicopathologic data of 489 patients with rectal cancer, retrospectively collected from a single institution between 2009 and 2013, were used to develop a predictive model for recurrence using the Cox regression. The model performance was validated on an independent cohort between 2015 and 2017 (N = 168). RESULTS: Out of 489 derivative patients, 103 showed recurrence after surgery. The prediction model was constructed with the following four significant predictors: distance from anal verge, MR-based extramural venous invasion, pathologic nodal stage, and perineural invasion (HR: 1.69, 2.09, 2.59, 2.29, respectively). Each factor was assigned a risk score corresponding to HR. The derivation and validation cohort were classified by sum of risk scores into 3 groups: low, intermediate, and high risk. Each of these groups showed significantly different recurrence rates (derivation cohort: 13.4%, 35.3%, 61.5 %; validation cohort: 6.2%, 23.7%, 64.7%). Our new model showed better performance in risk stratification, compared to recurrence rates of tumor node metastasis (TNM) staging in the validation cohort (stage I: 3.6%, II: 12%, III: 30.2%). The area under the receiver operating characteristic curve of the new prediction model was higher than TNM staging at 3-year recurrence in the validation cohort (0.853 vs. 0.731; p = .009). CONCLUSIONS: The new risk prediction model was strongly correlated with a recurrence rate after rectal cancer surgery and excellent for selection of high-risk group, who needs more active surveillance. KEY POINTS: • Multivariate analysis revealed four significant risk factors to be MR-based extramural venous invasion, perineural invasion, nodal metastasis, and the short distance from anal verge among the radiologic and clinicopathologic data. • Our new recurrence prediction model including radiologic data as well as clinicopathologic data showed high predictive performance of disease recurrence. • This model can be used as a comprehensive approach to evaluate individual prognosis and helpful for the selection of highly recurrent group who needs more active surveillance.

Hot snare polypectomy versus endoscopic mucosal resection for small colorectal polyps: a randomized controlled trial.

BACKGROUND AND AIMS: This study was performed to compare endoscopic mucosal resection (EMR) with hot snare polypectomy (HSP) in terms of the complete resection rate and the incidence of adverse events for resecting small (5-10 mm) colorectal polyps. METHODS: Small colorectal polyps (5-10 mm) with neoplastic features were randomly allocated to either the HSP or EMR group. A submucosal injection was performed prior to hot snaring in the EMR group only. Complete resection was defined as the absence of neoplastic tissue from two additional biopsies of the polypectomy site. R0 resection was defined as the absence of neoplastic tissue at the margin of the resected specimen. RESULTS: A total of 362 colon polyps from 272 patients were included, and 167 polyps in the HSP group and 155 polyps in the EMR group were analyzed. Between the polypectomy techniques, there was no significant difference in the complete resection rates, which were 96.4% (161/167) in the HSP group and 95.5% (148/155) in the EMR group (P = 0.67). The R0 resection rate in the HSP and EMR groups was significantly different, with 49.7% (83/167) and 74.8% (116/155), respectively (P < 0.001). There was no significant difference in the incidence of adverse events between the two groups. CONCLUSIONS: The complete resection rates for small (5-10 mm) polyps were not different between HSP and EMR. TRIAL REGISTRY: ClincialTrials.gov number NCT02239536.

Preferential Expression of Programmed Death Ligand 1 Protein in Tumor-Associated Macrophages and Its Potential Role in Immunotherapy for Hepatocellular Carcinoma.

A predictive biomarker of immune checkpoint inhibitor (ICI)-based treatments in hepatocellular carcinoma (HCC) has not been clearly demonstrated. In this study, we focused on the infiltration and programmed death ligand 1 (PD-L1) expression of tumor-associated macrophages (TAMs) in the tumor microenvironment of HCC. Immunohistochemistry demonstrated that PD-L1 was preferentially expressed on CD68+ macrophages in the tumor microenvironment of HCC, suggestive of its expression in TAMs rather than in T cells or tumor cells (P < 0.05). A co-culture experiment using activated T cells and M2 macrophages confirmed a significant increase in T cell functionality after the pretreatment of M2 macrophages with anti-PD-L1. Syngeneic mouse model experiments demonstrated that TAMs expressed PD-L1 and tumors treated with anti-PD-L1 showed smaller diameters than those treated with IgG. In these mice, anti-PD-L1 treatment increased activation markers in intratumoral CD8+ T cells and reduced the size of the TAM population. Regarding nivolumab-treated patients, three of eight patients responded to the anti-PD-1 treatment. The percentage of Ki-67-positive CD4+ and CD8+ T cells was higher in responders than non-responders after nivolumab. Overall, PD-L1 expression on TAMs may be targeted by immune-based HCC treatment, and ICI treatment results in the reinvigoration of exhausted CD8+ T cells in HCC.

Feasibility of fully automated classification of whole slide images based on deep learning.

Although microscopic analysis of tissue slides has been the basis for disease diagnosis for decades, intra- and inter-observer variabilities remain issues to be resolved. The recent introduction of digital scanners has allowed for using deep learning in the analysis of tissue images because many whole slide images (WSIs) are accessible to researchers. In the present study, we investigated the possibility of a deep learning-based, fully automated, computer-aided diagnosis system with WSIs from a stomach adenocarcinoma dataset. Three different convolutional neural network architectures were tested to determine the better architecture for tissue classifier. Each network was trained to classify small tissue patches into normal or tumor. Based on the patch-level classification, tumor probability heatmaps can be overlaid on tissue images. We observed three different tissue patterns, including clear normal, clear tumor and ambiguous cases. We suggest that longer inspection time can be assigned to ambiguous cases compared to clear normal cases, increasing the accuracy and efficiency of histopathologic diagnosis by pre-evaluating the status of the WSIs. When the classifier was tested with completely different WSI dataset, the performance was not optimal because of the different tissue preparation quality. By including a small amount of data from the new dataset for training, the performance for the new dataset was much enhanced. These results indicated that WSI dataset should include tissues prepared from many different preparation conditions to construct a generalized tissue classifier. Thus, multi-national/multi-center dataset should be built for the application of deep learning in the real world medical practice.

Heterogeneous nuclear ribonucleoprotein A1 promotes the expression of autophagy-related protein 6 in human colorectal cancer.

Autophagy, a lysosomal self-degradative process of cellular components, is essential for cellular homeostasis to response cellular stress and is tightly controlled by autophagy-related genes (ATGs). Autophagy-related gene 6 (ATG6, also known as Beclin-1 in human) is an essential factor regulating autophagy and apoptosis. RNA binding proteins (RBPs) regulate gene expression at the post-transcriptional level and their differential expression is linked to the pathogenesis of several human diseases. Here, we demonstrate the role of heterogeneous nuclear ribonucleoprotein A1 (hnRNPA1) as a novel factor regulating ATG6 expression. hnRNPA1 associates with the 3' untranslated region (3'UTR) of ATG6 mRNA and promotes its expression without significant changes at the mRNA level. Knockdown of hnRNPA1 decreases ATG6 expression, which is enhanced by the overexpression of hnRNPA1. Also, we show augmented expression of both hnRNPA1 and ATG6 in the colorectal cancer (CRC) tissues obtained from patients and demonstrate a positive correlation of their expression in CRC tissues. Our results suggest the potential role of hnRNPA1-mediated ATG6 regulation in the pathogenesis of CRC.

Efficacy of capecitabine and oxaliplatin versus S-1 as adjuvant chemotherapy in gastric cancer after D2 lymph node dissection according to lymph node ratio and N stage.

BACKGROUND: We sought to assess the prognostic significance of lymph node ratio (LNR) and N stage in patients undergoing D2 gastrectomy and adjuvant chemotherapy, S-1, and XELOX and to compare the efficacy of them according to LNRs and N stages to evaluate the clinical impact of using LNRs compared with using N staging. METHODS: Patients undergoing D2 gastrectomy with adequate lymph node dissection and adjuvant chemotherapy for stage II/III gastric cancer between Mar 2011 and Dec 2016 were analysed. Of the 477 patients enrolled, 331 received S-1 and 146 received XELOX. LNR groups were segregated as 0, 0-0.1, 0.1-0.25, and > 0.25 (LNR0, 1, 2, and 3, respectively). Propensity score matching (PSM) was used to minimise potential selection bias and compare DFS and OS stratified by LNRs and N stages in the two treatment groups. RESULTS: After PSM, the sample size of each group was 110 patients, and variables were well balanced. All patients had more than 15 examined lymph nodes (median 51, range 16~124). In multivariate analysis, LNR (> 0.25) and N stage (N3) showed independent prognostic value in OS and DFS, but LNR (> 0.25) showed better prognostic value. In subgroup analysis, the LNR3 group showed better 5-year DFS (20% vs 54%; HR 0.29; p = 0.004) and 5-year OS (26% vs 67%; HR 0.28; p = 0.020) in the XELOX group. The N3 group showed better 5-year DFS (38% vs 66%; HR 0.40; p = 0.004) and 5-year OS (47% vs 71%; HR 0.45; p = 0.019) in the XELOX group. Stage IIIC showed better 5-year DFS (22% vs 57%; HR 0.32; p = 0.004) and 5-year OS (27% vs 68%; HR 0.32; p = 0.009) in the XELOX group. The LNR3 group within N3 patients showed better 5-year DFS (21% vs 55%; HR 0.31; p = 0.004) and 5-year OS (27% vs 68%; HR 0.34; p = 0.018) in the XELOX group. CONCLUSIONS: LNR showed better prognostic value than N staging. LNR3, N3 and stage IIIC groups showed the superior efficacy of XELOX to that of S-1. And the LNR3 group within N3 patients showed more survival benefit from XELOX. LNR > 0.25, N3 stage and stage IIIC were the discriminant factors for selecting XELOX over S-1. TRIAL REGISTRATION: Not applicable (retrospective study).

Discordancy and changes in the pattern of programmed death ligand 1 expression before and after platinum-based chemotherapy in metastatic gastric cancer.

BACKGROUND: Our goal was to evaluate changes in PD-L1 expression in primary tumours of metastatic gastric cancer before and after chemotherapy. METHODS: We evaluated the PD-L1 expression of 72 patients with primary gastric cancer, before and after palliative first-line platinum-based chemotherapy, between January 2015 and March 2017. The PD-L1 ratio was defined as pre-chemotherapy PD-L1 expression divided by the post-chemotherapy PD-L1 expression. RESULTS: In 30 patients with PD-L1 negative pre-chemotherapy, 12 (40%) were positive post-chemotherapy; among the 42 patients with PD-L1 positive pre-chemotherapy, 24 (57.1%) were negative post-chemotherapy. The degree of PD-L1 expression decreased from 58.3% before chemotherapy to 41.7% after chemotherapy (P = 0.046). Among patients with complete response/partial response (CR/PR), the degree of PD-L1 expression decreased (P = 0.002), as well as PD-L1 positivity with statistical significance (P = 0.013) after chemotherapy, but not among patients with stable disease/progressive disease (SD/PD). Higher disease control rates (CR/PR/SD) were observed in patients with an elevated PD-L1 ratio (P = 0.043). Patients with a high PD-L1 ratio (> 1) were found to be associated with a better progression-free survival (HR 0.34, 95% CI 0.17-0.67, P = 0.002). CONCLUSIONS: PD-L1 expression can change during chemotherapy. Moreover, changes in patterns of PD-L1 expression might be associated with patient prognosis and response to chemotherapy.