Discordant PD-L1 results between 28-8 and 22C3 assays are associated with outcomes of gastric cancer patients treated with nivolumab plus chemotherapy.

BACKGROUND: We evaluated the concordance/discordance of PD-L1 staining results between the 28-8 and 22C3 assays and its impact on the efficacy outcomes of advanced gastric cancer patients treated with nivolumab plus chemotherapy. METHODS: This retrospective study involved 143 gastric cancer patients treated with first-line nivolumab plus chemotherapy whose PD-L1 results with both 28-8 and 22C3 assays were available. The concordance/discordance between these assays and the inter-observer variability were evaluated for PD-L1 combined positive score (CPS) positivity. Discordant PD-L1 results were analyzed regarding survival outcomes. RESULTS: The agreement rates and Cohen's kappa values between the 28-8 and 22C3 assays were 78.3% and 0.56 (for CPS ≥ 1), 81.8% and 0.60 (for CPS ≥ 5), and 88.8% and 0.66 (for CPS ≥ 10), respectively. Inter-observer variability, as represented by the intra-class correlation coefficient, was 0.89 and 0.88 for the 28-8 and 22C3 assays, respectively. With PD-L1 CPS ≥ 5 defined as positive, 35 (24.5%) and 82 (57.3%) had concordantly positive and negative results, respectively, between the 28-8 and 22C3 assays, whereas 26 (18.2%) had discordant results. Progression-free survival was shorter for those who exhibited negatively concordant PD-L1 results and discordant PD-L1 positivity between the 28-8 and 22C3 assays relative to those with positively concordant PD-L1 results (P = 0.013). CONCLUSION: PD-L1 assays by 28-8 and 22C3 showed suboptimal concordance, while inter-observer variability was not critical in advanced gastric cancer. Discordant PD-L1 results between 28-8 and 22C3 assays may be associated with unfavorable efficacy outcomes in patients treated with nivolumab plus chemotherapy.

Clinical implications of pediatric biliary intraepithelial neoplasia diagnosed from a choledochal cyst specimen.

BACKGROUND: Biliary intraepithelial neoplasia (BilIN), a noninvasive precursor of cholangiocarcinoma, can manifest malignant transformation. Since cholangiocarcinoma (CCA) may progress due to chronic inflammation in the bile ducts and gallbladder, choledochal cysts are considered a precursor to CCA. However, BilIN has rarely been reported in children, to date. METHODS: We reviewed medical records of patients (< 18 years of age, n = 329) who underwent choledochal cyst excision at Asan Medical Center from 2008 to 2022. BilIN was diagnosed in 15 patients. Subsequent analyses were performed of the demographics, surgical procedures, clinical course, and outcomes in these patients. Subgroup analysis and multivariate logistic regression test were performed to identify factors influencing BilIN occurrence. RESULTS: The mean age of the patients included in our study was 40.1 ± 47.6 months. In 15 patients, BilIN of various grades was diagnosed. Todani type I was prevalent in 80% of the patients. The median age at surgery was 17 months. During a mean follow-up of 63.3 ± 94.0 months, no adverse events such as stone formation in the remnant intrapancreatic common bile duct and intrahepatic duct or cholangiocarcinoma were observed, indicating a favorable outcome until now. CONCLUSIONS: The potential progression of choledochal cysts to BilIN in children was demonstrated. These results could underscore the importance of early and comprehensive excision of choledochal cysts, including resection margins for associated lesions and more thorough postoperative surveillance in patients with or at risk of BilIN.

Effectiveness of a healthy lifestyle program based on a mobile serious game for childhood cancer survivors: A quasi-randomized trial.

PURPOSE: This study aimed to develop and evaluate the effectiveness of a healthy lifestyle program based on a mobile serious game (HLP-MSG) to enhance the lifestyles of childhood cancer survivors (CCSs). METHODS: This program proceeded in two stages: development and evaluation, using a non-synchronized design with a quasi-randomized trial. The participants were CCSs aged 6-13 years whose treatment was terminated at least 12 months prior. Data were collected at baseline, and post-intervention, with a follow-up after four weeks using the Child Healthy Lifestyle Profile (CHLP). The experimental (n = 26) and control groups (n = 25) were compared. Data were analyzed using descriptive statistics, chi-squared tests, t-tests, and repeated-measures ANOVA. RESULTS: The HLP-MSG promoted a healthy lifestyle by solving 26 quests, including seven sub-elements (nutrition, exercise, hygiene, interpersonal relationships, stress management, meaning of life, and health responsibility). This study revealed significant differences in the interaction between measurement time and group assignment in the CHLP (p = .006) and physical activity (p = .013), one of the seven sub-dimensions. CONCLUSIONS: A healthy lifestyle program based on a mobile serious game is a feasible health education modality to enhance the physical, psychological, social, and spiritual health of CCSs. IMPLICATIONS TO PRACTICE: The findings add to scientific evidence on a mobile serious game for health education among CCSs. The HLP-MSG provides an evolutionary educational modality that can be delivered non-face-to-face to promote CCSs' continuous healthy behavior maintenance. Moreover, the HLP-MSG is adolescent-friendly and can be utilized as a healthcare tool for parents and children to cooperate.

Primary cardiac sarcomas: A clinicopathologic study in a single institution with 25 years of experience with an emphasis on MDM2 expression and adjuvant therapy for prognosis.

BACKGROUND: Primary cardiac sarcomas are rare and their clinicopathologic features are heterogeneous. Among them, particularly intimal sarcoma is a diagnostic challenge due to nonspecific histologic features. Recently, MDM2 amplification reported to be a characteristic genetic event in the intimal sarcoma. In this study, we aimed to identify the types and incidence of primary cardiac sarcomas that occurred over 25 years in tertiary medical institutions, and to find clinicopatholgical significance through reclassification of diagnoses using additional immunohistochemistry (IHC). METHODS: We reviewed the primary cardiac sarcoma cases between January 1993 and June 2018 at Asan Medical Center, South Korea, with their clinicopathologic findings, and reclassified the subtypes, especially using IHC for MDM2 and then, analyzed the significance of prognosis. RESULTS: Forty-eight (6.8%) cases of a primary cardiac sarcoma were retrieved. The tumors most frequently involved the right atrium (n = 25, 52.1%), and the most frequent tumor subtype was angiosarcoma (n = 23, 47.9%). Seven cases (53.8%) were newly reclassified as an intimal sarcoma by IHC for MDM2. Twenty-nine (60.4%) patients died of disease (mean, 19.8 months). Four patients underwent a heart transplantation and had a median survival of 26.8 months. This transplantation group tended to show good clinical outcomes in the earlier stages, but this was not statistically significant (p = 0.318). MDM2 positive intimal sarcoma showed the better overall survival (p = 0.003) than undifferentiated pleomorphic sarcoma. Adjuvant treatment is beneficial for patient survival (p < 0.001), particularly in angiosarcoma (p < 0.001), but not in intimal sarcoma (p = 0.154). CONCLUSION: Our study supports the use of adjuvant treatment in primary cardiac sarcoma, as it was associated with a significantly better overall survival rate. Further consideration of tumor histology may be important in determining the optimal use of adjuvant treatment for different types of sarcomas. Therefore, accurate diagnosis by MDM2 test is important condsidering patient's prognosis and treatment.

A Standardized Pathology Report for Gastric Cancer: 2nd Edition.

The first edition of 'A Standardized Pathology Report for Gastric Cancer' was initiated by the Gastrointestinal Pathology Study Group of the Korean Society of Pathologists and published 17 years ago. Since then, significant advances have been made in the pathologic diagnosis, molecular genetics, and management of gastric cancer (GC). To reflect those changes, a committee for publishing a second edition of the report was formed within the Gastrointestinal Pathology Study Group of the Korean Society of Pathologists. This second edition consists of two parts: standard data elements and conditional data elements. The standard data elements contain the basic pathologic findings and items necessary to predict the prognosis of GC patients, and they are adequate for routine surgical pathology service. Other diagnostic and prognostic factors relevant to adjuvant therapy, including molecular biomarkers, are classified as conditional data elements to allow each pathologist to selectively choose items appropriate to the environment in their institution. We trust that the standardized pathology report will be helpful for GC diagnosis and facilitate large-scale multidisciplinary collaborative studies.

A standardized pathology report for gastric cancer: 2nd edition.

The first edition of 'A Standardized Pathology Report for Gastric Cancer' was initiated by the Gastrointestinal Pathology Study Group of the Korean Society of Pathologists and published 17 years ago. Since then, significant advances have been made in the pathologic diagnosis, molecular genetics, and management of gastric cancer (GC). To reflect those changes, a committee for publishing a second edition of the report was formed within the Gastrointestinal Pathology Study Group of the Korean Society of Pathologists. This second edition consists of two parts: standard data elements and conditional data elements. The standard data elements contain the basic pathologic findings and items necessary to predict the prognosis of GC patients, and they are adequate for routine surgical pathology service. Other diagnostic and prognostic factors relevant to adjuvant therapy, including molecular biomarkers, are classified as conditional data elements to allow each pathologist to selectively choose items appropriate to the environment in their institution. We trust that the standardized pathology report will be helpful for GC diagnosis and facilitate large-scale multidisciplinary collaborative studies.

Diagnostic Assessment of Deep Learning Algorithms for Frozen Tissue Section Analysis in Women with Breast Cancer.

PURPOSE: Assessing the metastasis status of the sentinel lymph nodes (SLNs) for hematoxylin and eosin-stained frozen tissue sections by pathologists is an essential but tedious and time-consuming task that contributes to accurate breast cancer staging. This study aimed to review a challenge competition (HeLP 2019) for the development of automated solutions for classifying the metastasis status of breast cancer patients. Materials and Methods: A total of 524 digital slides were obtained from frozen SLN sections: 297 (56.7%) from Asan Medical Center (AMC) and 227 (43.4%) from Seoul National University Bundang Hospital (SNUBH), South Korea. The slides were divided into training, development, and validation sets, where the development set comprised slides from both institutions and training and validation set included slides from only AMC and SNUBH, respectively. The algorithms were assessed for area under the receiver operating characteristic curve (AUC) and measurement of the longest metastatic tumor diameter. The final total scores were calculated as the mean of the two metrics, and the three teams with AUC values greater than 0.500 were selected for review and analysis in this study. RESULTS: The top three teams showed AUC values of 0.891, 0.809, and 0.736 and major axis prediction scores of 0.525, 0.459, and 0.387 for the validation set. The major factor that lowered the diagnostic accuracy was micro-metastasis. CONCLUSION: In this challenge competition, accurate deep learning algorithms were developed that can be helpful for making a diagnosis on intraoperative SLN biopsy. The clinical utility of this approach was evaluated by including an external validation set from SNUBH.

Diagnostic Value of Nestin Expression in Adult Gliomas.

Introduction. Nestin, a type VI intermediate filament, is expressed in neuroepithelial cells during embryogenesis and has been expressed in various human tumors. Recent studies reported that the expression was associated with poor prognosis in brain tumors, but the results were inconclusive. In this study, we evaluated usefulness of nestin expression as a prognostic biomarker in consideration of IDH mutation and the World Health Organization (WHO) classification fifth edition. Methods. To investigate nestin expression, immunohistochemistry was performed on 92 adult brain gliomas using tissue microarrays. We analyzed the clinical characteristics and survival outcomes according to nestin expression and examined whether nestin expression alone affects the prognosis, independent of IDH mutation. Results. Sixty patients (65.2%) were nestin-positive (weak and strong). Nestin expression and intensity were significantly correlated with pathologic diagnosis and IDH mutation. The patients with high-grade gliomas showed a higher frequency and stronger intensity of nestin expression than those with low-grade gliomas (P < .001). The majority (93.6%) of gliomas with IDH mutation showed no expression or weak positivity. Multivariate Cox proportional hazard regression analysis for overall survival demonstrated that nestin expression (weak, hazard ratio [HR] 5.39, P = .036; strong, HR 8.43, P = .007) was an independent prognostic factor. Moreover, patients with nestin-expressing glioma showed shorter survival (P < .001). Conclusions. Nestin seems to be strongly expressed in the vast majority of glioblastomas, IDH-wildtype and rarely in IDH-mutant gliomas. Clear correlation between nestin expression and pathologic diagnosis makes an accurate patient diagnosis. Expression and intensity of nestin were significantly correlated with worse survival.

Survival Prediction after Curative Resection of Pancreatic Ductal Adenocarcinoma by Imaging-Based Intratumoral Necrosis.

We aimed to determine the histopathological characteristics and prognosis of curatively resected pancreatic ductal adenocarcinoma (PDAC) showing intratumoral necrosis on preoperative CT or MRI. This study consecutively included 102 patients who underwent upfront surgery with margin-negative resection from 2012 to 2020. All patients underwent both pancreatic CT and MRI within 1 month before surgery. Two radiologists independently assessed CT/MRI findings, including the presence of CT- and MRI-detected necrosis. Histopathological characteristics of PDACs according to CT or MRI detection of necrosis were evaluated. Disease-free survival (DFS) and overall survival (OS) were assessed by the Kaplan−Meier method and the Cox proportional hazards model. Among the 102 PDAC patients, 14 patients (13.7%) had CT-detected necrosis, and 16 patients (15.7%) had MRI-detected necrosis, of which 9 showed both CT- and MRI-detected necrosis. PDACs with CT- or MRI-detected necrosis demonstrated a significantly higher degree of histopathological necrosis than those without (p < 0.001). Multivariable analysis revealed that tumor size (hazard ratio [HR], 1.19; p = 0.040), tumor location (HR, 0.46; p = 0.009), and MRI-detected necrosis (HR, 2.64; p = 0.002) had independent associations with DFS. Only MRI-detected necrosis was significantly associated with OS (HR, 2.59; p = 0.004). Therefore, MRI-detected necrosis might be a potential imaging predictor of poor survival after curative resection of PDAC.

Molecular and immunohistochemical comparison between primary gastrointestinal mucosal melanomas and atypical gastrointestinal stromal tumors.

BACKGROUND: It is sometimes challenging to differentiate between gastrointestinal (GI) mucosal melanoma and gastrointestinal stromal tumors (GISTs) because they share some similarities in histological and molecular features. In this study, we investigated their clinicopathological and molecular features. METHODS: Four primary GI mucosal melanomas and 4 atypical GISTs resected from 2017 to 2019 were included. Electronic medical records and histology slides were reviewed and, if needed, immunohistochemical stainings were additionally done. Somatic mutations were analyzed using whole exome sequencing (WES) with tumors and matched normal tissues. RESULTS: By immunohistochemistry, GISTs were positive for CD117 (4 of 4), DOG1 (4 of 4), and CD34 (2 of 4). In contrast, melanomas were positive for CD117 (2 of 4), HMB-45 (4 of 4), Melan-A (4 of 4), and S100 protein (3 of 4). Using WES, the KIT mutation was detected in 2 of 4 GISTs and 1 of 4 melanomas, all of which showed strong CD117 immunoreactivity. PDGFRA mutation was detected in 2 of 4 GISTs but 0 of 4 melanomas. TP53, NF1, RB1, TERT, and KMT2D mutations were detected in 1 of 4 melanomas but 0 of 4 GISTs. A patient with KIT-mutated melanoma was treated with Gleevec and has remained disease-free for 1 year. CONCLUSIONS: GISTs and GI melanomas have a wide histopathological spectrum. Appropriate ancillary studies can be helpful for proper diagnosis and optimal treatment.

Identification of intratumoral fluid-containing area by magnetic resonance imaging to predict prognosis in patients with pancreatic ductal adenocarcinoma after curative resection.

OBJECTIVES: To compare the prognosis of pancreatic ductal adenocarcinoma (PDAC) after curative resection according to the type of intratumoral fluid-containing area identified on MRI. METHODS: This retrospective study included 112 consecutive patients who underwent upfront surgery with margin-negative resection between 2012 and 2019. All patients underwent MRI within 1 month before surgery. Three radiologists independently assessed the MRI findings, determined whether intratumoral fluid-containing areas were present, and classified all intratumoral fluid-containing areas by type (i.e., imaging necrosis or neoplastic mucin cysts). Recurrence-free survival (RFS) and overall survival (OS) were evaluated by the Kaplan-Meier method and the Cox proportional hazards model. Histopathological differences according to the type of intratumoral fluid-containing area were assessed. RESULTS: Of the 112 PDAC patients, intratumoral fluid-containing areas were identified on MRI in 33 (29.5%), among which 18 were classified as imaging necrosis and 15 as neoplastic mucin cysts. PDAC patients with imaging necrosis demonstrated significantly shorter RFS (mean 6.1 months versus 47.3 months; p < .001) and OS (18.4 months versus 55.0 months, p = .001) than those with neoplastic mucin cysts. Multivariable analysis showed that only the type of intratumoral fluid-containing area was significantly associated with RFS (hazard ratio, 2.25 and 0.38; p = .009 and p = .046 for imaging necrosis and neoplastic mucin cysts, respectively). PDAC with imaging necrosis had more frequent histological necrosis, more aggressive tumor differentiation, and higher tumor cellularity than PDAC with neoplastic mucin cysts (p ≤ .02). CONCLUSION: The detection and discrimination of intratumoral fluid-containing areas on preoperative MRI may be useful in predicting the prognosis of PDAC patients after curative resection. KEY POINTS: • Pancreatic ductal adenocarcinoma (PDAC) patients with imaging necrosis demonstrated significantly shorter survival than those with neoplastic mucin cysts after curative resection. • Multivariable analysis showed that only the type of intratumoral fluid-containing area identified on MRI was significantly associated with recurrence-free survival. • PDAC with imaging necrosis had more frequent histological necrosis, more aggressive tumor differentiation, and higher tumor cellularity than PDAC with neoplastic mucin cysts.

High incidence of late anastomosis leakage in patients for rectal cancer after neoadjuvant chemoradiotherapy: A comparative study.

PURPOSE: The purpose of this study was to investigate the clinical features and risk factors of late anastomotic leakage (AL) in a homogeneous cohort with elective sphincter-sparing surgery (SSS) with ileostomy after neoadjuvant chemoradiotherapy (nCRT) for rectal cancer. METHODS: Data from a total of 359 patients who underwent elective rectal cancer surgery between Jan 2017 and May 2020 were retrospectively reviewed. Patients were classified into early and late AL groups, referring to onset of AL occurring within or after 30 post-operative days, respectively. We analyzed clinical, pathological, and inflammatory features of both AL and risk factors of stoma reversal failure and late AL. RESULTS: A total of 85 patients with SSS with ileostomy after nCRT were classified into 8 (9.4%) patients of early AL and 16 (18.8%) of late AL. Unlike early AL patients, late AL group showed lower neutrophil-lymphocyte ratio (NLR) (P < 0.001) and did not need an invasive intervention at the time of diagnosis. 50% (5/10) patients needed reformation of ileostomy. (P = 0.048) Failure of stoma reversal is associated with advanced stages, high NLR ratio (≥3), and inflammatory lesions seen around anastomosis in radiologic findings, which was confirmed as the risk factor of late AL. CONCLUSION: Late AL, with different clinical features, showed a higher incidence than early AL in patients who underwent surgery after nCRT and also had a higher stoma reformation rate. Careful evaluation using laboratory and radiological findings before an ileostomy closure is performed to prevent late AL.

The Association of Estrogen Receptor Activity, Interferon Signaling, and MHC Class I Expression in Breast Cancer.

PURPOSE: The expression of major histocompatibility complex class I (MHC I) has previously been reported to be negatively associated with estrogen receptor (ER) expression. Furthermore, MHC I expression, level of tumor-infiltrating lymphocytes (TILs), and expression of interferon (IFN) mediator MxA are positively associated with one another in human breast cancers. This study aimed to investigate the mechanisms of association of MHC I with ER and IFN signaling. MATERIALS AND METHODS: The human leukocyte antigen (HLA)-ABC protein expression was analyzed in breast cancer cell lines. The expressions of HLA-A and MxA mRNAs were analyzed in MCF-7 cells in Gene Expression Omnibus (GEO) data. ER and HLA-ABC expressions, Ki-67 labeling index and TIL levels in tumor tissue were also analyzed in ER+/ human epidermal growth factor receptor 2 (HER2)- breast cancer patients who randomly received either neoadjuvant chemotherapy or estrogen modulator treatment followed by resection. RESULTS: HLA-ABC protein expression was decreased after ß-estradiol treatment or hESR-GFP transfection and increased after fulvestrant or IFN-γ treatment in cell lines. In GEO data, HLA-A and MxA expression was increased after ESR1 shRNA transfection. In patients, ER Allred score was significantly lower and the HLA-ABC expression, TIL levels, and Ki-67 were significantly higher in the estrogen modulator treated group than the chemotherapy treated group. CONCLUSION: MHC I expression and TIL levels might be affected by ER pathway modulation and IFN treatment. Further studies elucidating the mechanism of MHC I regulation could suggest a way to boost TIL influx in cancer in a clinical setting.

Prediction of genetic alterations from gastric cancer histopathology images using a fully automated deep learning approach.

BACKGROUND: Studies correlating specific genetic mutations and treatment response are ongoing to establish an effective treatment strategy for gastric cancer (GC). To facilitate this research, a cost- and time-effective method to analyze the mutational status is necessary. Deep learning (DL) has been successfully applied to analyze hematoxylin and eosin (H and E)-stained tissue slide images. AIM: To test the feasibility of DL-based classifiers for the frequently occurring mutations from the H and E-stained GC tissue whole slide images (WSIs). METHODS: From the GC dataset of The Cancer Genome Atlas (TCGA-STAD), wild-type/mutation classifiers for CDH1, ERBB2, KRAS, PIK3CA, and TP53 genes were trained on 360 × 360-pixel patches of tissue images. RESULTS: The area under the curve (AUC) for the receiver operating characteristic (ROC) curves ranged from 0.727 to 0.862 for the TCGA frozen WSIs and 0.661 to 0.858 for the TCGA formalin-fixed paraffin-embedded (FFPE) WSIs. The performance of the classifier can be improved by adding new FFPE WSI training dataset from our institute. The classifiers trained for mutation prediction in colorectal cancer completely failed to predict the mutational status in GC, indicating that DL-based mutation classifiers are incompatible between different cancers. CONCLUSION: This study concluded that DL could predict genetic mutations in H and E-stained tissue slides when they are trained with appropriate tissue data.

A New Human Leukocyte Antigen Typing Algorithm Combined With Currently Available Genotyping Tools Based on Next-Generation Sequencing Data and Guidelines to Select the Most Likely Human Leukocyte Antigen Genotype.

Background: High-precision human leukocyte antigen (HLA) genotyping is crucial for anti-cancer immunotherapy, but existing tools predicting HLA genotypes using next-generation sequencing (NGS) data are insufficiently accurate. Materials and Methods: We compared availability, accuracy, correction score, and complementary ratio of eight HLA genotyping tools (OptiType, HLA-HD, PHLAT, seq2HLA, arcasHLA, HLAscan, HLA*LA, and Kourami) using 1,005 cases from the 1000 Genomes Project data. We created a new HLA-genotyping algorithm combining tools based on the precision and the accuracy of tools' combinations. Then, we assessed the new algorithm's performance in 39 in-house samples with normal whole-exome sequencing (WES) data and polymerase chain reaction-sequencing-based typing (PCR-SBT) results. Results: Regardless of the type of tool, the calls presented by more than six tools concordantly showed high accuracy and precision. The accuracy of the group with at least six concordant calls was 100% (97/97) in HLA-A, 98.2% (112/114) in HLA-B, 97.3% (142/146) in HLA-C. The precision of the group with at least six concordant calls was over 98% in HLA-ABC. We additionally calculated the accuracy of the combination tools considering the complementary ratio of each tool and the accuracy of each tool, and the accuracy was over 98% in all groups with six or more concordant calls. We created a new algorithm that matches the above results. It was to select the HLA type if more than six out of eight tools presented a matched type. Otherwise, determine the HLA type experimentally through PCR-SBT. When we applied the new algorithm to 39 in-house cases, there were more than six matching calls in all HLA-A, B, and C, and the accuracy of these concordant calls was 100%. Conclusions: HLA genotyping accuracy using NGS data could be increased by combining the current HLA genotyping tools. This new algorithm could also be useful for preliminary screening to decide whether to perform an additional PCR-based experimental method instead of using tools with NGS data.

Deep Learning for Automatic Subclassification of Gastric Carcinoma Using Whole-Slide Histopathology Images.

Histomorphologic types of gastric cancer (GC) have significant prognostic values that should be considered during treatment planning. Because the thorough quantitative review of a tissue slide is a laborious task for pathologists, deep learning (DL) can be a useful tool to support pathologic workflow. In the present study, a fully automated approach was applied to distinguish differentiated/undifferentiated and non-mucinous/mucinous tumor types in GC tissue whole-slide images from The Cancer Genome Atlas (TCGA) stomach adenocarcinoma dataset (TCGA-STAD). By classifying small patches of tissue images into differentiated/undifferentiated and non-mucinous/mucinous tumor tissues, the relative proportion of GC tissue subtypes can be easily quantified. Furthermore, the distribution of different tissue subtypes can be clearly visualized. The patch-level areas under the curves for the receiver operating characteristic curves for the differentiated/undifferentiated and non-mucinous/mucinous classifiers were 0.932 and 0.979, respectively. We also validated the classifiers on our own GC datasets and confirmed that the generalizability of the classifiers is excellent. The results indicate that the DL-based tissue classifier could be a useful tool for the quantitative analysis of cancer tissue slides. By combining DL-based classifiers for various molecular and morphologic variations in tissue slides, the heterogeneity of tumor tissues can be unveiled more efficiently.

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.

Enhancing the landscape of colorectal cancer using targeted deep sequencing.

Targeted next-generation sequencing (NGS) technology detects specific mutations that can provide treatment opportunities for colorectal cancer (CRC) patients. We included 145 CRC patients who underwent surgery. We analyzed the mutation frequencies of common actionable genes and their association with clinicopathological characteristics and oncologic outcomes using targeted NGS. Approximately 97.9% (142) of patients showed somatic mutations. Frequent mutations were observed in TP53 (70%), APC (60%), and KRAS (49%). TP53 mutations were significantly linked to higher overall stage (p = 0.038) and lower disease-free survival (DFS) (p = 0.039). ATM mutation was significantly associated with higher tumor stage (p = 0.012) and shorter overall survival (OS) (p = 0.041). Stage 3 and 4 patients with ATM mutations (p = 0.023) had shorter OS, and FBXW7 mutation was significantly associated with shorter DFS (p = 0.002). However, the OS of patients with or without TP53, RAS, APC, PIK3CA, and SMAD4 mutations did not differ significantly (p = 0.59, 0.72, 0.059, 0.25, and 0.12, respectively). Similarly, the DFS between patients with RAS, APC, PIK3CA, and SMAD4 mutations and those with wild-type were not statistically different (p = 0.3, 0.79, 0.13, and 0.59, respectively). In multivariate Cox regression analysis, ATM mutation was an independent biomarker for poor prognosis of OS (p = 0.043). A comprehensive analysis of the molecular markers for CRC can provide insights into the mechanisms underlying disease progression and help optimize a personalized therapy.

Effectiveness of transfer learning for enhancing tumor classification with a convolutional neural network on frozen sections.

Fast and accurate confirmation of metastasis on the frozen tissue section of intraoperative sentinel lymph node biopsy is an essential tool for critical surgical decisions. However, accurate diagnosis by pathologists is difficult within the time limitations. Training a robust and accurate deep learning model is also difficult owing to the limited number of frozen datasets with high quality labels. To overcome these issues, we validated the effectiveness of transfer learning from CAMELYON16 to improve performance of the convolutional neural network (CNN)-based classification model on our frozen dataset (N = 297) from Asan Medical Center (AMC). Among the 297 whole slide images (WSIs), 157 and 40 WSIs were used to train deep learning models with different dataset ratios at 2, 4, 8, 20, 40, and 100%. The remaining, i.e., 100 WSIs, were used to validate model performance in terms of patch- and slide-level classification. An additional 228 WSIs from Seoul National University Bundang Hospital (SNUBH) were used as an external validation. Three initial weights, i.e., scratch-based (random initialization), ImageNet-based, and CAMELYON16-based models were used to validate their effectiveness in external validation. In the patch-level classification results on the AMC dataset, CAMELYON16-based models trained with a small dataset (up to 40%, i.e., 62 WSIs) showed a significantly higher area under the curve (AUC) of 0.929 than those of the scratch- and ImageNet-based models at 0.897 and 0.919, respectively, while CAMELYON16-based and ImageNet-based models trained with 100% of the training dataset showed comparable AUCs at 0.944 and 0.943, respectively. For the external validation, CAMELYON16-based models showed higher AUCs than those of the scratch- and ImageNet-based models. Model performance for slide feasibility of the transfer learning to enhance model performance was validated in the case of frozen section datasets with limited numbers.

Prediction of clinically actionable genetic alterations from colorectal cancer histopathology images using deep learning.

BACKGROUND: Identifying genetic mutations in cancer patients have been increasingly important because distinctive mutational patterns can be very informative to determine the optimal therapeutic strategy. Recent studies have shown that deep learning-based molecular cancer subtyping can be performed directly from the standard hematoxylin and eosin (H&E) sections in diverse tumors including colorectal cancers (CRCs). Since H&E-stained tissue slides are ubiquitously available, mutation prediction with the pathology images from cancers can be a time- and cost-effective complementary method for personalized treatment. AIM: To predict the frequently occurring actionable mutations from the H&E-stained CRC whole-slide images (WSIs) with deep learning-based classifiers. METHODS: A total of 629 CRC patients from The Cancer Genome Atlas (TCGA-COAD and TCGA-READ) and 142 CRC patients from Seoul St. Mary Hospital (SMH) were included. Based on the mutation frequency in TCGA and SMH datasets, we chose APC, KRAS, PIK3CA, SMAD4, and TP53 genes for the study. The classifiers were trained with 360 × 360 pixel patches of tissue images. The receiver operating characteristic (ROC) curves and area under the curves (AUCs) for all the classifiers were presented. RESULTS: The AUCs for ROC curves ranged from 0.693 to 0.809 for the TCGA frozen WSIs and from 0.645 to 0.783 for the TCGA formalin-fixed paraffin-embedded WSIs. The prediction performance can be enhanced with the expansion of datasets. When the classifiers were trained with both TCGA and SMH data, the prediction performance was improved. CONCLUSION: APC, KRAS, PIK3CA, SMAD4, and TP53 mutations can be predicted from H&E pathology images using deep learning-based classifiers, demonstrating the potential for deep learning-based mutation prediction in the CRC tissue slides.