http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
심층신경망 네트워크의 잠재변수를 이용한 Glioblastoma 환자군 생존분석 알고리즘
조환호(Hwan-ho Cho),박현진(Hyunjin Park) 대한전자공학회 2020 대한전자공학회 학술대회 Vol.2020 No.8
Glioblastoma is the most lethal primary brain tumor with a median survival of about 12 months. Recently, deep neural networks show excellent performance in image classification or object detection by learning the abstract features of images themselves. In this study, we proposed an algorithm for survival analysis of patients with glioblastoma through latent variables obtained by feeding medical images into a pre-trained deep learning network. In the validation step, our algorithm showed a correlation of 0.3545 with pvalue 0.0017 for survival day regression, 0.6806 area under the curve value of receiver operating characteristics curve for long-, short-term survival group classification, and risk group stratification showed p-value of 0.0035 in the log-rank test for Kaplan-Meier analysis.
생성적 대립 신경망을 이용한 교모세포종 환자의 조영 증강 T1 강조 자기공명영상 합성
김은진(Eunjin Kim),조환호(Hwan-ho Cho),권준모(Junmo Kwon),박현진(Hyunjin Park) 대한전자공학회 2020 대한전자공학회 학술대회 Vol.2020 No.8
Glioblastoma shows diverse shape and heterogeneous characteristics, thus magnetic resonance imaging (MRI) images from multiple sequences are required for an accurate diagnosis. Especially, T1 contrast—enhanced (T1CE) imaging is one of the most important imaging modalities for the detection and diagnosis of glioblastoma. However, patients are exposed to potentially dangerous injection of gadolinium—based contrast agents. Generative Adversarial Network (GAN) learns the generator and discriminator simultaneously to generate the synthesized data that approximates the real data. In this paper, we proposed the GAN based image—to—image translation algorithm to synthesize T1CE MRI images from T1 MRI images. The performance of the proposed model was evaluated using similarity measurements including mean squared error (MSE), peak signal—to—noise ratio (PSNR). As a result, the mean values of MSE and PSNR from the test data set were 14.857 and 39.817 dB improving existing studies.
인셉션 기반 멀티모달 융합네트워크와 뇌영상을 이용한 뇌교종의 등급 구분
정승완(Seung-Wan Jeong),조환호(Hwan-ho Cho),권준모(Junmo Kwon),박현진(Hyunjin Park) 대한전자공학회 2020 대한전자공학회 학술대회 Vol.2020 No.8
Glioma is a brain tumor that occurs in the neuroglia cell in the brain. It is important to classify the grades of glioma because the prognosis and method of treatment vary greatly depending on the grades of glioma. In this paper, we propose a fusion network—based inception network that effectively uses information from multi—modal brain images to classify glioma grades. The Brain Tumor Segmentation Challenge 2017 (BraTS 2017) dataset was used to evaluate the network. The network received multi—modal images as input and was validated using the 5—fold cross—validation. As a result, our method achieved AUC value 0.9412 on average on the test fold, which is 0.005 to 0.01 higher than other competing methods.