다중혈관 관상동맥 환자에서 y-문합을 이용하여 양쪽 내흉동맥만을 사용한 우회술의 조기 성적

Probability Distributions.- Linear Models for Regression.- Linear Models for Classification.- Neural Networks.- Kernel Methods.- Sparse Kernel Machines.- Graphical Models.- Mixture Models and EM.- Approximate Inference.- Sampling Methods.- Continuous Latent Variables.- Sequential Data.- Combining Models.

Pattern Recognition and Machine Learning

“Unpaired Image-to-Image Translation using Cycle-Consistent Adversarial Networks”の学習報告

Digital processing of speech signals

QUITE apart from the academic consideration that vital and medical statistics now form an obligatory part of the education of students seeking the University of London's diploma in public health, the demand for information about the methods of vital and medical statistics is increasing. The most casual reader of the newspapers is now aware that population problems are of serious practical importance and that the publications of the General Register Office cannot be ignored. It is scarcely an exaggeration to say that the recently issued preliminary report on the census of 1931 is one of the most sensational documents which has appeared for years, and that he who reads it intelligently will understand what is meant by saying that civilisation is in the melting pot. An Introduction to Medical Statistics. By Hilda M. Woods William T. Russell. Pp. x + 125. (London: P. S. King and Son, Ltd., 1931.) 7s. 6d.

/pdf/an-introduction-to-medical-statistics-58gm2cuxfv.pdf

An Introduction to Medical Statistics

In this work, we propose a wavelet-based technique for embedding medical data in a medical image. In the spectral domain, the patient data are embedded into the wavelet coefficients of the host image. A diagnostic distortion measure (DDM) has been defined to measure the visible distortions between the original image and the embedded image. The performance of the DDM has been compared with the standard PSNR characteristics. The results show that DDM not only captures the distortions for different quantities of embedded data but also can quantify the differences when the same data are embedded at different subbands. The embedment of data into the mid and high frequency subbands of the host image show lower values of DDM and higher values of PSNR.

Wavelet transform domain data embedding in a medical image

Deep learning algorithms can offer a reliable automated interpretation of retinal optical coherence tomography (OCT) images to assist clinicians in disease diagnosis and management. However, retinal image processing presents pertinent obstacles such as the struggle of large-scale data acquisition and high cost of annotation. To address this, we have developed a data-efficient semisupervised generative adversarial network based classifier for automated diagnosis with limited labeled data. The framework consists of a generator and a discriminator. The adversarial learning between them assists in building a generalizable classifier to predict progressive retinal diseases like age-related macular degeneration and diabetic macular edema. Experimental results on clinical-grade OCT images show an overall improvement of more than 10% in accuracy, compared to the state-of-the-art methods.

A Data-Efficient Approach for Automated Classification of OCT Images Using Generative Adversarial Network

Spikes such as QRS complex in ECG, epileptic seizures in EEG, fine crackles in vesicular sound and glottal closure instants in voiced sound are of diagnostic importance. Various methods of spike detection use the amplitude and frequency characteristics of the spikes. Because of the high frequency content, the spikes appear in the error signal when a linear prediction filtering scheme is used. The authors use the method of midprediction filtering for the detection of the spikes. In this method, the present sample is predicted as a weighted average of p recent past and p immediate future samples. The symmetrical nature of midprediction causes the spikes to appear in the error signal with their original basewidths. This can help in improving the reliability of spike detection, as both the amplitude and the duration of the spike can be considered as decision making parameters. It is observed that the high frequency gain of the midprediction filter is higher compared to the high frequency gain of the LPC or endprediction filter. As a result, this method works better than linear prediction for the detection of spikes.

Spike detection in biomedical signals using midprediction filter

Myocardial infarction is a coronary artery ailment, and it is characterized by the changes in the morphological features such as the shape of T-wave, Q-wave and ST-segment of ECG signal. In clinical standard, it is a challenging problem to diagnose MI pathology using 12-lead ECG and vectorcardiogram (VCG). VCG has the advantage to record the heart electrical activities in three orthogonal planes (frontal, sagittal and transverse). This paper proposes a new method for automated detection or grading of MI pathology from vectorcardiogram (VCG) signals. The method uses relevance vector machine (RVM) classifier and the multiscale features of VCG signal for MI detection. The multiscale analysis of VCG signal is performed using dual-tree complex wavelet transform. The diagnostic features such as the complex wavelet sub-band (CWS) $$L_{1}$$
 -Norm (CWS $$L_{1}$$
 -norm) and the complex wavelet entropy (CWE) are evaluated from the sub-band complex wavelet coefficients of each orthogonal lead of VCG. The RVM classifier is considered to evaluate the performance of the combination of the CWS $$L_{1}$$
 -norm and the CWE features of VCG. Three different kernel functions such as Gaussian, bubble and Cauchy are used for RVM. The results show that the RVM classifier with Gaussian kernel function has an average accuracy, an average sensitivity and an average specificity values of 99.80, 99.67 and 99.90%, respectively. The performance of RVM classifier is compared with the existing methods for detection of MI from VCG and 12-lead ECG signals.

Detection of myocardial infarction from vectorcardiogram using relevance vector machine

This work explores the effect of breathiness component on speech under stress. The breathiness component in a speech signal can be estimated using different features such as period perturbation quotient (PPQ), amplitude perturbation quotient (APQ), harmonic to noise ratio (HNR), glottal to noise excitation ratio (GNER), harmonic energy (HE), harmonic energy of residue (HER) and harmonic to signal ratio (HSR). Statistical analysis of these features shows that they have different mean and variance values for speech under stress. The performance of breathiness features is evaluated using Hidden Markov Model (HMM) for classification of speech under stress. The results show that the breathiness features successfully characterize the speech under stress. The performance of breathiness features is compared with the MFCC feature. Finally, a speech under stress classification method is proposed with the combination of breathiness and MFCC features. In terms of classification rates, the proposed combined feature outperforms the MFCC feature.

Samarendra Dandapat

Papers

Wavelet transform domain data embedding in a medical image

A Data-Efficient Approach for Automated Classification of OCT Images Using Generative Adversarial Network

Spike detection in biomedical signals using midprediction filter

Detection of myocardial infarction from vectorcardiogram using relevance vector machine

A novel breathiness feature for analysis and classification of speech under stress