• Journal of Korea Multimedia Society
• /
• v.16 no.3
• /
• pp.269-280
• /
• 2013

In this paper, we suggested the possibility of using an efficient classifier for the dependency analysis of retinal disease. First, we analyzed the classification performance and the prediction accuracy of GBN (General Bayesian Network), GBN with reduced features by Markov Blanket and TAN (Tree-Augmented Naive Bayesian Network) among the various bayesian networks. And then, for the first time, we applied TAN showing high performance to the dependency analysis of the clinical data of retinal disease. As a result of this analysis, it showed applicability in the diagnosis and the prediction of prognosis of retinal disease.

• Asian Pacific Journal of Cancer Prevention
• /
• v.15 no.1
• /
• pp.101-105
• /
• 2014

Background and Purpose: Human epididymis protein 4 (HE4) has been suggested to be a novel biomarker of epithelial ovarian cancer (EOC). The present study aimed to evaluate and compare HE4 with the commonly used marker, carbohydrate antigen 125 (CA125), in prediction and therapy-monitoring of EOC. Patients and Methods: Serum HE4 concentrations from 123 ovarian cancer patients and 174 controls were measured by Roche electrochemiluminescent immunoassay (ECLIA). Risk of ovarian malignancy algorithm (ROMA) values were calculated and assessed. In addition, the prospects of HE4 detection for therapy-monitoring were evaluated in EOC patients. Results: The ROMA score could classify patients into high- and low-risk groups with malignancy. Indeed, lower serum HE4 was significantly associated with successful surgical therapy. Specifically, 38 patients with EOC exhibited a greater decline of HE4 compared with CA125. In contrast, elevation of HE4 better predicted recurrence (of 46, 11 patients developed recurrence, and with it increased HE4 serum concentrations) and a poor prognosis than CA125. Conclusions: This study suggests that serum HE4 levels are closely associated with outcome of surgical therapy and disease prognosis in Chinese EOC patients.

• Journal of Oral Medicine and Pain
• /
• v.23 no.3
• /
• pp.263-270
• /
• 1998

Considering various factors contributing oral malodor, the accurate prediction of prognosis is very important to both clinician and patients. The present study has been performed to invetigate the relationship between treatment effeicacy and pre-treatment volatile sulfur compounds (VSC) level. Ninety patients were divided into three groups, A(<150ppb), B(150< <200ppb), and C(>200ppb) groups, according to pre-treatment VSC level detected by Halimeter, and each group included 30 patients. Routine therapeutic measures for oral were provided to each patient which consisted of oral prophylaxis, tooth brushing and flossing instruction, tongue scraping by proper device, and gargling of 0.25% ZnCl2 Solution. The group with high pre-treatment VSC level (>150ppb) showed significant reduction of VSC level at 1 and 3 weeks after. However, the group with low pre-treatment VSC level (<150ppb) did not show any significant reduction during the experimental periods. Collectively, the results suggested that patients with high pre-treatment VSC level show better prognosis.

• KIPS Transactions on Software and Data Engineering
• /
• v.8 no.10
• /
• pp.397-402
• /
• 2019

Accurately predicting cancer prognosis to provide appropriate treatment strategies for patients is one of the critical challenges in bioinformatics. Many researches have suggested machine learning models to predict patients' outcomes based on their gene expression data. Gene expression data is high-dimensional numerical data containing about 17,000 genes, so traditional researches used feature selection or dimensionality reduction approaches to elevate the performance of prognostic prediction models. These approaches, however, have an issue of making it difficult for the predictive models to grasp any biological interaction between the selected genes because feature selection and model training stages are performed independently. In this paper, we propose a novel two-dimensional image formatting approach for gene expression data to achieve feature selection and prognostic prediction effectively. Node2Vec is exploited to integrate biological interaction network and gene expression data and a convolutional neural network learns the integrated two-dimensional gene expression image data and predicts cancer prognosis. We evaluated our proposed model through double cross-validation and confirmed superior prognostic prediction accuracy to traditional machine learning models based on raw gene expression data. As our proposed approach is able to improve prediction models without loss of information caused by feature selection steps, we expect this will contribute to development of personalized medicine.

• Journal of Yeungnam Medical Science
• /
• v.38 no.4
• /
• pp.318-325
• /
• 2021

Background: The diagnosis and prediction of prognosis are important in patients with sepsis, and presepsin is helpful. In this study, we aimed to examine the usefulness of presepsin in predicting the prognosis of sepsis in Korea. Methods: Patients diagnosed with sepsis according to the sepsis-3 criteria were recruited into the study and classified into surviving and non-surviving groups based on in-hospital mortality. A total of 153 patients (32 and 121 patients with sepsis and septic shock, respectively) were included from July 2019 to August 2020. Results: Among the 153 patients with sepsis, 91 and 62 were in the survivor and non-survivor groups, respectively. Presepsin (p=0.004) and lactate (p=0.003) levels and the sequential organ failure assessment (SOFA) score (p<0.001) were higher in the non-survivor group. Receiver operating characteristic curve analysis revealed poor performances of presepsin and lactate in predicting the prognosis of sepsis (presepsin: area under the curve [AUC]=0.656, p=0.001; lactate: AUC=0.646, p=0.003). The SOFA score showed the best performance, with the highest AUC value (AUC=0.751, p<0.001). The prognostic cutoff point for presepsin was 1,176 pg/mL. Presepsin levels higher than 1,176 pg/mL (odds ratio [OR], 3.352; p<0.001), higher lactate levels (OR, 1.203; p=0.003), and higher SOFA score (OR, 1.249; p<0.001) were risk factors for in-hospital mortality. Conclusion: Presepsin levels were higher in non-survivors than in survivors. Thus, presepsin may be a valuable biomarker in predicting the prognosis of sepsis.

• Journal of Gastric Cancer
• /
• v.20 no.4
• /
• pp.385-394
• /
• 2020

Purpose: Patients with gastric cancer who receive neoadjuvant therapy are staged before treatment (cStage) and after treatment (ypStage). We aimed to compare the prognostic reliability of cStage and ypStage, alone and in combination. Materials and Methods: Data for all patients who received neoadjuvant therapy followed by surgery for gastric adenocarcinoma from 2004 to 2015 were extracted from the National Cancer Database. Kaplan-Meier (KM)curves were used to model overall survival based on cStage alone, ypStage alone, cStage stratified by ypStage, and ypStage stratified by cStage. P-values were generated to summarize the differences in KM curves. The discriminatory power of survival prediction was examined using Harrell's C-statistics. Results: We included 8,977 patients in the analysis. As expected, increasing cStage and ypStage were associated with worse survival. The discriminatory prognostic power provided by cStage was poor (C-statistic 0.548), while that provided by ypStage was moderate (C-statistic 0.634). Within each cStage, the addition of ypStage information significantly altered the prognosis (P<0.0001 within cStages I-IV). However, for each ypStage, the addition of cStage information generally did not alter the prognosis (P=0.2874, 0.027, 0.061, 0.049, and 0.007 within ypStages 0-IV, respectively). The discriminatory prognostic power provided by the combination of cStage and ypStage was similar to that of ypStage alone (C-statistic 0.636 vs. 0.634). Conclusions: The cStage is unreliable for prognosis, and ypStage is moderately reliable. Combining cStage and ypStage does not improve the discriminatory prognostic power provided by ypStage alone. A ypStage-based prognosis is minimally affected by the initial cStage.

• Journal of Information Processing Systems
• /
• v.19 no.5
• /
• pp.673-687
• /
• 2023

Convolutional neural network-based deep learning technology is the most commonly used in image identification, but it requires large-scale data for training. Therefore, application in specific fields in which data acquisition is limited, such as in the military, may be challenging. In particular, the identification of ground weapon systems is a very important mission, and high identification accuracy is required. Accordingly, various studies have been conducted to achieve high performance using small-scale data. Among them, the ensemble method, which achieves excellent performance through the prediction average of the pre-trained models, is the most representative method; however, it requires considerable time and effort to find the optimal combination of ensemble models. In addition, there is a performance limitation in the prediction results obtained by using an ensemble method. Furthermore, it is difficult to obtain the ensemble effect using models with imbalanced classification accuracies. In this paper, we propose a transfer learning-based feature fusion technique for heterogeneous models that extracts and fuses features of pre-trained heterogeneous models and finally, fine-tunes hyperparameters of the fully connected layer to improve the classification accuracy. The experimental results of this study indicate that it is possible to overcome the limitations of the existing ensemble methods by improving the classification accuracy through feature fusion between heterogeneous models based on transfer learning.

• Asian Pacific Journal of Cancer Prevention
• /
• v.13 no.8
• /
• pp.4113-4117
• /
• 2012

Breast cancer is a prevalent heterogeneous malignant disease. Gene expression profiling by DNA microarray can classify breast tumors into five different molecular subtypes: luminal A, luminal B, HER-2, basal and normal-like which have differing prognosis. Recently it has been shown that immunohistochemistry (IHC) markers including estrogen receptor (ER), progesterone receptor (PR) and human epidermal growth factor receptor 2 (Her2), can divide tumors to main subtypes: luminal A (ER+; PR+/-; HER-2-), luminal B (ER+;PR+/-; HER-2+), basal-like (ER-;PR-;HER2-) and Her2+ (ER-; PR-; HER-2+). Some subtypes such as basal-like subtype have been characterized by poor prognosis and reduced overall survival. Due to the importance of the ER signaling pathway in mammary cell proliferation; it appears that epigenetic changes in the $ER{\alpha}$ gene as a central component of this pathway, may contribute to prognostic prediction. Thus this study aimed to clarify the correlation of different IHC-based subtypes of breast tumors with $ER{\alpha}$ methylation in Iranian breast cancer patients. For this purpose one hundred fresh breast tumors obtained by surgical resection underwent DNA extraction for assessment of their ER methylation status by methylation specific PCR (MSP). These tumors were classified into main subtypes according to IHC markers and data were collected on pathological features of the patients. $ER{\alpha}$ methylation was found in 25 of 28 (89.3%) basal tumors, 21 of 24 (87.5%) Her2+ tumors, 18 of 34 (52.9%) luminal A tumors and 7 of 14 (50%) luminal B tumors. A strong correlation was found between $ER{\alpha}$ methylation and poor prognosis tumor subtypes (basal and Her2+) in patients (P<0.001). Our findings show that $ER{\alpha}$ methylation is correlated with poor prognosis subtypes of breast tumors in Iranian patients and may play an important role in pathogenesis of the more aggressive breast tumors.

• Asian Pacific Journal of Cancer Prevention
• /
• v.15 no.1
• /
• pp.11-16
• /
• 2014

Accurate diagnosis and proper monitoring of cancer patients remain important obstacles for successful cancer treatment. The search for cancer biomarkers can aid in more accurate prediction of clinical outcome and may also reveal novel predictive factors and therapeutic targets. One such prognostic marker seems to be FOXA1. Many studies have shown that FOXA1 is strongly expressed in a vast majority of cancers, including breast cancer, in which high expression is associated with a good prognosis. In this review, we summarize the role of this transcription factor in the development and prognosis of breast cancer in the hope of providing insights into utility of FOXA1 as a novel biomarker.

• Asian Pacific Journal of Cancer Prevention
• /
• v.16 no.7
• /
• pp.2601-2611
• /
• 2015

Prostate cancer, with a lifetime prevalence of one in six men, is the second cause of malignancy-related death and the most prevalent cancer in men in many countries. Nowadays, prostate cancer diagnosis is often based on the use of biomarkers, especially prostate-specific antigen (PSA) which can result in enhanced detection at earlier stage and decreasing in the number of metastatic patients. However, because of the low specificity of PSA, unnecessary biopsies and mistaken diagnoses frequently occur. Prostate cancer has various features so prognosis following diagnosis is greatly variable. There is a requirement for new prognostic biomarkers, particularly to differentiate between inactive and aggressive forms of disease, to improve clinical management of prostate cancer. Research continues into finding additional markers that may allow this goal to be attained. We here selected a group of candidate biomarkers including PSA, PSA velocity, percentage free PSA, $TGF{\beta}1$, AMACR, chromogranin A, IL-6, IGFBPs, PSCA, biomarkers related to cell cycle regulation, apoptosis, PTEN, androgen receptor, cellular adhesion and angiogenesis, and also prognostic biomarkers with Genomic tests for discussion. This provides an outline of biomarkers that are presently of prognostic interest in prostate cancer investigation.