Artificial intelligence and computational pathology.

Data processing and learning has become a spearhead for the advancement of medicine, with pathology and laboratory medicine has no exception. The incorporation of scientific research through clinical informatics, including genomics, proteomics, bioinformatics, and biostatistics, into clinical practice unlocks innovative approaches for patient care. Computational pathology is burgeoning subspecialty in pathology that promises a better-integrated solution to whole-slide images, multi-omics data, and clinical informatics. However, computational pathology faces several challenges, including the ability to integrate raw data from different sources, limitation of hardware processing capacity, and a lack of specific training programs, as well as issues on ethics and larger societal acceptable practices that are still solidifying. The establishment of the entire industry of computational pathology requires far-reaching changes of the three essential elements connecting patients and doctors: the local laboratory, the scan center, and the central cloud hub/portal for data processing and retrieval. Computational pathology, unlocked through information integration and advanced digital communication networks, has the potential to improve clinical workflow efficiency, diagnostic quality, and ultimately create personalized diagnosis and treatment plans for patients. This review describes clinical perspectives and discusses the statistical methods, clinical applications, potential obstacles, and future directions of computational pathology.

Low-level parental somatic mosaic SNVs in exomes from a large cohort of trios with diverse suspected Mendelian conditions.

PURPOSE: The goal of this study was to assess the scale of low-level parental mosaicism in exome sequencing (ES) databases. METHODS: We analyzed approximately 2000 family trio ES data sets from the Baylor-Hopkins Center for Mendelian Genomics (BHCMG) and Baylor Genetics (BG). Among apparent de novo single-nucleotide variants identified in the affected probands, we selected rare unique variants with variant allele fraction (VAF) between 30% and 70% in the probands and lower than 10% in one of the parents. RESULTS: Of 102 candidate mosaic variants validated using amplicon-based next-generation sequencing, droplet digital polymerase chain reaction, or blocker displacement amplification, 27 (26.4%) were confirmed to be low- (VAF between 1% and 10%) or very low (VAF <1%) level mosaic. Detection precision in parental samples with two or more alternate reads was 63.6% (BHCMG) and 43.6% (BG). In nine investigated individuals, we observed variability of mosaic ratios among blood, saliva, fibroblast, buccal, hair, and urine samples. CONCLUSION: Our computational pipeline enables robust discrimination between true and false positive candidate mosaic variants and efficient detection of low-level mosaicism in ES samples. We confirm that the presence of two or more alternate reads in the parental sample is a reliable predictor of low-level parental somatic mosaicism.

BET N-terminal bromodomain inhibition selectively blocks Th17 cell differentiation and ameliorates colitis in mice.

T-helper 17 (Th17) cells have important functions in adaptor immunity and have also been implicated in inflammatory disorders. The bromodomain and extraterminal domain (BET) family proteins regulate gene transcription during lineage-specific differentiation of naïve CD4+ T cells to produce mature T-helper cells. Inhibition of acetyl-lysine binding of the BET proteins by pan-BET bromodomain (BrD) inhibitors, such as JQ1, broadly affects differentiation of Th17, Th1, and Th2 cells that have distinct immune functions, thus limiting their therapeutic potential. Whether these BET proteins represent viable new epigenetic drug targets for inflammatory disorders has remained an unanswered question. In this study, we report that selective inhibition of the first bromodomain of BET proteins with our newly designed small molecule MS402 inhibits primarily Th17 cell differentiation with a little or almost no effect on Th1 or Th2 and Treg cells. MS402 preferentially renders Brd4 binding to Th17 signature gene loci over those of housekeeping genes and reduces Brd4 recruitment of p-TEFb to phosphorylate and activate RNA polymerase II for transcription elongation. We further show that MS402 prevents and ameliorates T-cell transfer-induced colitis in mice by blocking Th17 cell overdevelopment. Thus, selective pharmacological modulation of individual bromodomains likely represents a strategy for treatment of inflammatory bowel diseases.

ATRX-mediated chromatin association of histone variant macroH2A1 regulates α-globin expression.

The histone variant macroH2A generally associates with transcriptionally inert chromatin; however, the factors that regulate its chromatin incorporation remain elusive. Here, we identify the SWI/SNF helicase ATRX (α-thalassemia/MR, X-linked) as a novel macroH2A-interacting protein. Unlike its role in assisting H3.3 chromatin deposition, ATRX acts as a negative regulator of macroH2A's chromatin association. In human erythroleukemic cells deficient for ATRX, macroH2A accumulates at the HBA gene cluster on the subtelomere of chromosome 16, coinciding with the loss of α-globin expression. Collectively, our results implicate deregulation of macroH2A's distribution as a contributing factor to the α-thalassemia phenotype of ATRX syndrome.

A recurrent 8 bp frameshifting indel in FOXF1 defines a novel mutation hotspot associated with alveolar capillary dysplasia with misalignment of pulmonary veins.

Alveolar capillary dysplasia with misalignment of pulmonary veins (ACDMPV) is a rare lethal lung developmental disease. Affected infants manifest with severe respiratory distress and refractory pulmonary hypertension and uniformly die in the first month of life. Heterozygous point mutations or copy-number variant deletions involving FOXF1 and/or its upstream lung-specific enhancer on 16q24.1 have been identified in the vast majority of ACDMPV patients. We have previously described two unrelated families with a de novo pathogenic frameshift variant c.691_698del (p.Ala231Argfs*61) in the exon 1 of FOXF1. Here, we present a third unrelated ACDMPV family with the same de novo variant and propose that a direct tandem repeat of eight consecutive nucleotides GCGGCGGC within the ~4 kb CpG island in FOXF1 exon 1 is a novel mutation hotspot causative for ACDMPV.

Repeated craniotomies for intracranial tumors: is the risk increased? Pooled analysis of two prospective, institutional registries of complications and outcomes.

PURPOSE: Deciding whether to re-operate patients with intracranial tumor recurrence or remnant is challenging, as the data on safety of repeated procedures is limited. This study set out to evaluate the risks for morbidity, mortality, and complications after repeated operations, and to compare those to primary operations. METHODS: Retrospective observational two-center study on consecutive patients undergoing microsurgical tumor resection. The data derived from independent, prospective institutional registries. The primary endpoint was morbidity at 3 months (M3), defined as significant decrease on the Karnofsky Performance Scale (KPS). Secondary endpoints were mortality, rate and severity of complications according to the Clavien-Dindo Grade (CDG). RESULTS: 463/2403 (19.3%) were repeated procedures. Morbidity at M3 occurred in n = 290 patients (12.1%). In univariable analysis, patients undergoing repeated surgery were 98% as likely as patients undergoing primary surgery to experience morbidity (OR 0.98, 95% CI 0.72-1.34, p = 0.889). In multivariable analysis adjusted for age, sex, tumor size, histology and posterior fossa location, the relationship remained stable (aOR 1.25, 95% CI 0.90-1.73, p = 0.186). Mortality was n = 10 (0.4%) at discharge and n = 95 (4.0%) at M3, without group differences. At least one complication occurred in n = 855, and the rate (35.5% vs. 35.9%, p = 0.892) and severity (CDG; p = 0.520) was similar after primary and repeated procedures. Results were reproduced in subgroup analyses for meningiomas, gliomas and cerebral metastases. CONCLUSIONS: Repeated surgery for intracranial tumors does not increase the risk of morbidity. Mortality, and both the rate and severity of complications are comparable to primary operations. This information is of value for patient counseling and the informed consent process.

Protein expression profiles and clinicopathologic characteristics associate with gastric cancer survival.

BACKGROUND: Prognosis remains one of most crucial determinants of gastric cancer (GC) treatment, but current methods do not predict prognosis accurately. Identification of additional biomarkers is urgently required to identify patients at risk of poor prognoses. METHODS: Tissue microarrays were used to measure expression of nine GC-associated proteins in GC tissue and normal gastric tissue samples. Hierarchical cluster analysis of microarray data and feature selection for factors associated with survival were performed. Based on these data, prognostic scoring models were established to predict clinical outcomes. Finally, ingenuity pathway analysis (IPA) was used to identify a biological GC network. RESULTS: Eight proteins were upregulated in GC tissues versus normal gastric tissues. Hierarchical cluster analysis and feature selection showed that overall survival was worse in cyclin dependent kinase (CDK)2, Akt1, X-linked inhibitor of apoptosis protein (XIAP), Notch4, and phosphorylated (p)-protein kinase C (PKC) α/ß2 immunopositive patients than in patients that were immunonegative for these proteins. Risk score models based on these five proteins and clinicopathological characteristics were established to determine prognoses of GC patients. These proteins were found to be involved in cancer related-signaling pathways and upstream regulators were identified. CONCLUSION: This study identified proteins that can be used as clinical biomarkers and established a risk score model based on these proteins and clinicopathological characteristics to assess GC prognosis.

The influence of preoperative dependency on mortality, functional recovery and complications after microsurgical resection of intracranial tumors.

INTRODUCTION: The decision whether to operate on patients with intracranial tumors is complex and influenced by patient-specific factors, including the preoperative functional status. This work assesses the risks for mortality and complications, and post-operative recovery in functionally dependent patients undergoing microsurgical resection of intracranial tumors. METHODS: Observational two-center study, analyzing institutional registry data. Dependency was defined as admission Karnofsky Performance Scale (KPS) of ≤ 50. The primary endpoint was in-hospital mortality. Secondary endpoints were rate and type [Clavien-Dindo grade (CDG)] of complications, as well as postoperative change in KPS until the 3-month follow-up (M3). RESULTS: Of n = 1951 patients, n = 98 (5.0%) were dependent. Mortality rates were 2.0% for dependent and 0.4% for independent patients (p = 0.018). In univariable analysis, dependent patients were more likely than independent patients to die in hospital (OR 5.49, 95% CI 1.12-26.8, p = 0.035). In a multivariable model, the effect was slightly attenuated (OR 4.75, 95% CI 0.91-24.7, p = 0.064). Dependent patients tended to experience more postoperative complications. They were more likely to suffer from a severe complication (CDG 4 and 5; OR 3.55, 95% CI 1.49-8.46, p = 0.004). In 40.8 and 52.4% of cases, dependent patients regained functional independence at discharge and M3, respectively. CONCLUSIONS: In operated patients with intracranial tumors presenting functionally dependent at admission, the risk for in-hospital mortality and complications is elevated. However, if conducted successfully, surgery may lead to regain of independence in every second patient within 3 months.

A hepatic stellate cell gene expression signature associated with outcomes in hepatitis C cirrhosis and hepatocellular carcinoma after curative resection.

OBJECTIVE: We used an informatics approach to identify and validate genes whose expression is unique to hepatic stellate cells and assessed the prognostic capability of their expression in cirrhosis. DESIGN: We defined a hepatic stellate cell gene signature by comparing stellate, immune and hepatic transcriptome profiles. We then created a prognostic index using a combination of hepatic stellate cell signature expression and clinical variables. This signature was derived in a retrospective-prospective cohort of hepatitis C-related early-stage cirrhosis (prognostic index derivation set) and validated in an independent retrospective cohort of patients with postresection hepatocellular carcinoma (HCC). We then examined the association between hepatic stellate cell signature expression and decompensation, HCC development, progression of Child-Pugh class and survival. RESULTS: The 122-gene hepatic stellate cell signature consists of genes encoding extracellular matrix proteins and developmental factors and correlates with the extent of fibrosis in human, mouse and rat datasets. Importantly, association of clinical prognostic variables with overall survival was improved by adding the signature; we used these results to define a prognostic index in the derivation set. In the validation set, the same prognostic index was associated with overall survival. The prognostic index was associated with decompensation, HCC and progression of Child-Pugh class in the derivation set, and HCC recurrence in the validation set. CONCLUSIONS: This work highlights the unique transcriptional niche of stellate cells, and identifies potential stellate cell targets for tracking, targeting and isolation. Hepatic stellate cell signature expression may identify patients with HCV cirrhosis or postresection HCC with poor prognosis.

RSK1 activation promotes invasion in nodular melanoma.

The two major melanoma histologic subtypes, superficial spreading and nodular melanomas, differ in their speed of dermal invasion but converge biologically once they invade and metastasize. Herein, we tested the hypothesis that distinct molecular alterations arising in primary melanoma cells might persist as these tumors progress to invasion and metastasis. Ribosomal protein S6 kinase, 90 kDa, polypeptide 1 (RSK1; official name RPS6KA1) was significantly hyperactivated in human melanoma lines and metastatic tissues derived from nodular compared with superficial spreading melanoma. RSK1 was constitutively phosphorylated at Ser-380 in nodular but not superficial spreading melanoma and did not directly correlate with BRAF or MEK activation. Nodular melanoma cells were more sensitive to RSK1 inhibition using siRNA and the pharmacological inhibitor BI-D1870 compared with superficial spreading cells. Gene expression microarray analyses revealed that RSK1 orchestrated a program of gene expression that promoted cell motility and invasion. Differential overexpression of the prometastatic matrix metalloproteinase 8 and tissue inhibitor of metalloproteinases 1 in metastatic nodular compared with metastatic superficial spreading melanoma was observed. Finally, using an in vivo zebrafish model, constitutive RSK1 activation increased melanoma invasion. Together, these data reveal a novel role for activated RSK1 in the progression of nodular melanoma and suggest that melanoma originating from different histologic subtypes may be biologically distinct and that these differences are maintained as the tumors invade and metastasize.

Induction and contribution of beta platelet-derived growth factor signalling by hepatic stellate cells to liver regeneration after partial hepatectomy in mice.

BACKGROUND & AIMS: Hepatic stellate cells (HSCs) activate during injury to orchestrate the liver's inflammatory and fibrogenic responses. A critical feature of HSC activation is the rapid induction of beta platelet-derived growth factor (ß-PDGFR), which drives cellular fibrogenesis and proliferation; in contrast, normal liver has minimal ß-PDGFR expression. While the role of ß-PDGFR is well established in liver injury, its expression and contribution during liver regeneration are unknown. The aim of this study was to determine whether ß-PDGFR is induced during liver regeneration following partial hepatectomy (pHx), and to define its contribution to the regenerative response. METHODS: Control mice or animals with HSC-specific ß-PDGFR-depletion underwent two-thirds pHx followed by assessment of hepatocyte proliferation and expression of ß-PDGFR. RNA-sequencing from whole liver tissue of both groups after pHx was used to uncover pathways regulated by ß-PDGFR signalling in HSCs. RESULTS: Beta platelet-derived growth factor expression on HSCs was up-regulated within 24 h following pHx in control mice, whereas absence of ß-PDGFR blunted the expansion of HSCs. Mice lacking ß-PDGFR displayed prolonged increases of transaminase levels within 72 h following pHx. Hepatocyte proliferation was impaired within the first 24 h based on Ki-67 and PCNA expression in ß-PDGFR-deficient mice. This was associated with dysregulated growth in the ß-PDGFR-deficient mice based on RNAseq with pathway analysis, and real-time quantitative PCR, which demonstrated reduced expression of Hgf, Igfbp1, Mapk and Il-6. CONCLUSIONS: Beta platelet-derived growth factor is induced in HSCs following surgical pHx and its deletion in HSCs leads to prolonged liver injury. However, there is no significant difference in liver regeneration.

IL-1ß induces IL-6 production in retinal Müller cells predominantly through the activation of p38 MAPK/NF-κB signaling pathway.

IL-6 plays an important role in various inflammatory ocular diseases, including diabetic retinopathy. Müller cells are the major source of inflammatory mediators, including IL-6, in the retina. However, the mechanism of regulating IL-6 production in these cells remains unclear. Examination of signaling pathways in human retinal Müller cells (MIO-M1 cell line) cultured with IL-1ß, TNF-α, IL-6, IL-8, VEGF, IFN-γ, glucose or mannitol showed that IL-1ß was the most potent stimulator of IL-6 production. In addition, IL-1 ß also increased NF-κB p50 protein level and phosphorylation of p38 MAPK, ERK1/2 and c-Jun. Induction of IL-6 production by IL-1ß was significantly reduced by addition of p38 MAPK (SB203580), MEK1/2 (U0126) or NF-κB (BAY11-7082) inhibitors, with the highest effect being observed with SB203580. To explore the specific elements in IL-6 promoter responsible for IL-1ß-induction of IL-6 expression, a series of plasmids bearing various IL-6 promoter mutations were transiently expressed in MIO-MI cells cultured in the presence or absence of IL-1ß (10ng/ml) and/or SB203580 (10µM). Results showed that IL-6 promoter activity of the parent pIL-6-Luc651 was significantly enhanced by IL-1ß, but the level was significantly attenuated by SB203580. Furthermore, the IL-6 promoter activity was also reduced upon deletion of NF-κB, AP-1 or C/EBP binding sites, with NF-κB deletion being the greatest. These results are the first demonstration that IL-1ß induces IL-6 production in Müller cells by activation of IL-6 promoter activity predominantly through the p38 MAPK/NF-κB pathway.

Protein profiling of Helicobacter pylori-associated gastric cancer.

Helicobacter pylori infection is an initiating factor in the development of gastric cancer. Gastric cancer can be divided into two groups on the basis of H. pylori serological status; seropositive H. pylori status predicts favorable prognosis in patients with gastric cancer. By using the protein pathway array, we identified 20 differentially expressed proteins in primary gastric cancer tissues between the H. pylori-seropositive and H. pylori-seronegative groups. Our results indicate that both brassinosteroid insensitive 1-associated kinase 1 and calpastatin are favorable prognostic factors in H. pylori-seropositive gastric cancer patients. In contrast, dachshund homolog 1 is a favorable prognostic factor in H. pylori-seronegative gastric cancer patients. Different signaling pathways were found to be altered between H. pylori-seropositive and H. pylori-seronegative gastric cancer, which may account for the different tumorigenesis and outcomes between these two subsets of patients.

Signaling proteins and pathways affected by flavonoids in leukemia cells.

Flavonoids are a class of plant secondary metabolites that are found ubiquitously in plants and in the human diet. Our objective is to investigate the antiproliferative effects of flavonoids (baicalein, luteolin, genistein, apigenin, scutellarin, galangin, chrysin, and naringenin) toward leukemia cells (HL-60, NB4, U937, K562, Jurkat) as well as the relationship between their antileukemic potencies and molecular structures. At the proteomic level, we evaluate the effects of different flavonoids on the expression levels of various proteins using Protein Pathway Array (PPA) technology. Our results showed a dose-dependent cytotoxicity of flavonoids toward various types of leukemia cells. The results of PPA illustrated that flavonoids, such as baicalein, genistein, and scutellarin affected different proteins in different leukemia cell lines. Cell cycle regulatory proteins, such as CDK4, CDK6, Cyclin D1, Cyclin B1, p-CDC2, and p-RB were affected in different leukemia cells. Furthermore, we found that baicalein suppresses CDK4 and activates p-ERK in most leukemia cells; genistein mainly affects CDK4, p-ERK, p-CDC2, while scutellarin dysregulated the proteins, cell division control protein 42, Notch4, and XIAP. Collectively, a wide variety of dysregulation of key signaling proteins related to apoptosis and cell-cycle regulation contributes to the antileukemic properties of these flavonoids.

High prevalence of HPV59 in cytologically abnormal cervical samples.

BACKGROUND: Cervical cancer is the second most common form of cancer among women. There are over 100 different types of human papillomavirus (HPV), 40 of which are frequently detected in anogenital mucosa. HPV is the primary etiological agent of cervical cancer and is present predominantly in cervical cancers. Thirteen commonly recognized high-risk genotypes have oncogenic potential. The most common high-risk HPV (hrHPV) genotypes in cervical cancer are HPV16 and HPV18, which have the greatest malignancies. The objective of this study was to determine the distribution of hrHPV types in patient samples received at the Mount Sinai Medical Center for routine cytology and HPV testing. In addition, the study compared the sensitivity of the Hybrid Capture 2 (HC2) and Roche Linear Array HPV Genotyping assays. METHODS: Cytology specimens with abnormal diagnoses were tested for HPV by the Roche Linear Array HPV Genotyping and Digene Hybrid Capture 2 molecular systems. RESULTS: The Roche Linear Array HPV Genotyping Assay was more sensitive than HC2. Additionally, specimens exhibited higher rates of HPV16, HPV51, and HPV59 infections than HPV16 and HPV18 infections. CONCLUSION: The results demonstrate that geographical distribution of HPV genotypes may play an important role in clinical management of HPV infection, particularly when treating cervical dysplasia and recommending HPV vaccination.

BRD4 inhibitor inhibits colorectal cancer growth and metastasis.

Post-translational modifications have been identified to be of great importance in cancers and lysine acetylation, which can attract the multifunctional transcription factor BRD4, has been identified as a potential therapeutic target. In this paper, we identify that BRD4 has an important role in colorectal cancer; and that its inhibition substantially wipes out tumor cells. Treatment with inhibitor MS417 potently affects cancer cells, although such effects were not always outright necrosis or apoptosis. We report that BRD4 inhibition also limits distal metastasis by regulating several key proteins in the progression of epithelial-to-mesenchymal transition (EMT). This effect of BRD4 inhibitor is demonstrated via liver metastasis in animal model as well as migration and invasion experiments in vitro. Together, our results demonstrate a new application of BRD4 inhibitor that may be of clinical use by virtue of its ability to limit metastasis while also being tumorcidal.

STK11 domain XI mutations: candidate genetic drivers leading to the development of dysplastic polyps in Peutz-Jeghers syndrome.

Peutz-Jeghers syndrome (PJS) is a rare hereditary disorder resulting from mutations in serine/threonine kinase 11 (STK11) and characterized by gastrointestinal (GI) hamartomatous polyps, mucocutaneous pigmentation, and an increased risk for specific cancers. Little is known about the genetic implications of specific STK11 mutations with regard to their role in dysplastic and malignant transformation of GI polyps. Peripheral blood genomic DNA samples from 116 Chinese PJS patients from 52 unrelated families were investigated for STK11 mutations. Genotype-phenotype correlations were investigated. The mutation detection rate was 67.3% (51.9% point mutations, 15.4% large deletions). Fourteen out of the 25 point mutations identified were novel. Nearly one-third of all mutations, 8/27 (29.6%), were in exon 7, the shortest out of the nine exons. Strikingly, mutations affecting protein kinase domain XI, encoded in part by exon 7, correlated with a 90% (9/10) incidence of GI polyp dysplasia. In contrast, only two out of 17 (11.8%) nondomain XI mutations were linked to polyp dysplasia (P = 0.0001). The extent of the association between dysplasia and the development of GI-related cancers is currently unknown but our results highlight a novel STK11 genotype-phenotype association as the basis for future genetic counseling and basic research studies.

Anti-tumor activity of the X-linked inhibitor of apoptosis (XIAP) inhibitor embelin in gastric cancer cells.

This study investigated the anticancer effects of embelin in human gastric cancer cells and the underlying molecular mechanisms. Gastric cancer cells were treated with embelin and 5-FU for methyl-thiazolyl-tetrazolium bromide cell viability assay and flow cytometric detection of cell viability and apoptosis. Protein pathway array (PPA) and Western blot were used to investigate differentially expressed proteins in embelin-treated gastric cancer cells. Embelin reduced gastric cancer cell viability, induced apoptosis, and enhanced 5-FU antitumor activity in gastric cancer cells. Mechanistically, embelin induced cell cycle arrest at the S and G2/M phases. Molecularly, embelin downregulated expression of X-linked inhibitor of apoptosis and cell cycle-regulatory proteins, such as CDK1, CDC25B, CDC25C, cyclinB1, and CDK2. PPA analysis showed that embelin modulated several pathways that are associated with cell growth and apoptosis, such as PI3K/AKT, JAK/STAT, p38 MAPK, and p53. The data from the current study implied that reduction of gastric cancer cell viability after treatment with embelin was through cell cycle arrest at the S and G2/M phases and apoptosis.

Protein predictive signatures for lymph node metastasis of gastric cancer.

Lymph node status remains one of most crucial indicators of gastric cancer prognosis and treatment planning. Current imaging methods have limited accuracy in predicting lymph node metastasis. We sought to identify protein markers in primary gastric cancer and to define a risk model to predict lymph node metastasis. The Protein Pathway Array (PPA) (initial selection) and Western blot (confirmation) were used to assess the protein expression in a total of 190 freshly frozen gastric cancer samples. The protein expression levels were compared between samples with lymph node metastasis (n = 73) and those without lymph node metastasis (n = 57) using PPA. There were 27 proteins differentially expressed between lymph node positive samples and lymph node negative samples. Five proteins (Factor XIII B, TFIIH p89, ADAM8, COX-2 and CUL-1) were identified as independent predictors of lymph node metastasis. Together with vascular/lymphatic invasion status, a risk score model was established to determine the risk of lymph node metastasis for each individual gastric cancer patient. The ability of this model to predict lymph node metastasis was further confirmed in a second cohort of gastric cancer patients (33 with and 27 without lymph node metastasis) using Western blot. This study indicated that some proteins differentially expressed in gastric cancer can be selected as clinically useful biomarkers. The risk score model is useful for determining patients' risk of lymph node metastasis and prognosis.