Improving Computer-Aided Thoracic Disease Diagnosis through Comparative Analysis Using Chest X-ray Images Taken at Different Times.

Medical professionals in thoracic medicine routinely analyze chest X-ray images, often comparing pairs of images taken at different times to detect lesions or anomalies in patients. This research aims to design a computer-aided diagnosis system that enhances the efficiency of thoracic physicians in comparing and diagnosing X-ray images, ultimately reducing misjudgments. The proposed system encompasses four key components: segmentation, alignment, comparison, and classification of lung X-ray images. Utilizing a public NIH Chest X-ray14 dataset and a local dataset gathered by the Chiayi Christian Hospital in Taiwan, the efficacy of both the traditional methods and deep-learning methods were compared. Experimental results indicate that, in both the segmentation and alignment stages, the deep-learning method outperforms the traditional method, achieving higher average IoU, detection rates, and significantly reduced processing time. In the comparison stage, we designed nonlinear transfer functions to highlight the differences between pre- and post-images through heat maps. In the classification stage, single-input and dual-input network architectures were proposed. The inclusion of difference information in single-input networks enhances AUC by approximately 1%, and dual-input networks achieve a 1.2-1.4% AUC increase, underscoring the importance of difference images in lung disease identification and classification based on chest X-ray images. While the proposed system is still in its early stages and far from clinical application, the results demonstrate potential steps forward in the development of a comprehensive computer-aided diagnostic system for comparative analysis of chest X-ray images.

Emotion Recognition Based on Electroencephalogram Using Semi-supervised Generative Adversarial Network.

This paper proposes the use of Semi-supervised Generative Adversarial Network (SGAN) to take advantage of the large amount of unlabeled electroencephalogram (EEG) spectrogram data in improving the classifier's accuracy in emotion recognition. The use of SGAN led the discriminator network to not just learn in a supervised fashion from the small amount of labeled data to distinguish among the different target classes, but also make use of the true unlabeled data to distinguish them from the synthetic ones generated by the generator network. This additional ability to distinguish true and fake samples forces the network to focus only on features that are present on a true sample to distinguish the classes, thereby improving generalization and overall accuracy. An ablation study is devised, where the SGAN classifier is compared to a mere discriminator network without the GAN architecture. The 80% : 20% validation method was employed to classify the EEG spectrogram of 50 participants gathered by Kaohsiung Medical University into two emotion labels in the valence dimension: positive and negative. The proposed method achieved an accuracy of 84.83% given only 50% labeled data, which is not just better than the baseline discriminator network which achieved 83.5% accuracy, but is also better than many previous studies at accuracies around 78%. This demonstrates the ability of SGAN in improving discriminator network's accuracy by training it to also distinguish between the unlabeled true sample and synthetic data.Clinical Relevance- The use of EEG in emotion recognition has seen growing interest due to its ease of access. However, the large amount of labeled data required to train an accurate model has been the limiting factor as databases in the area of emotion recognition with EEG is still relatively small. This paper proposes the use of SGAN to allow using large amount of unlabeled EEG data to improve the recognition rate.

Affective Computing Based on Morphological Features of Photoplethysmography for Patients with Hypertension.

Negative and positive emotions are the risk and protective factors for the cause and prognosis of hypertension. This study aimed to use five photoplethysmography (PPG) waveform indices and affective computing (AC) to discriminate the emotional states in patients with hypertension. Forty-three patients with essential hypertension were measured for blood pressure and PPG signals under baseline and four emotional conditions (neutral, anger, happiness, and sadness), and the PPG signals were transformed into the mean standard deviation of five PPG waveform indices. A support vector machine was used as a classifier. The performance of the classifier was verified by using resubstitution and six-fold cross-validation (CV) methods. Feature selectors, including full search and genetic algorithm (GA), were used to select effective feature combinations. Traditional statistical analyses only differentiated between the emotional states and baseline, whereas AC achieved 100% accuracy in distinguishing between the emotional states and baseline by using the resubstitution method. AC showed high accuracy rates when used with 10 waveform features in distinguishing the records into two, three, and four classes by applying a six-fold CV. The GA feature selector further boosted the accuracy to 78.97%, 74.22%, and 67.35% in two-, three-, and four-class differentiation, respectively. The proposed AC achieved high accuracy in categorizing PPG records into distinct emotional states with features extracted from only five waveform indices. The results demonstrated the effectiveness of the five indices and the proposed AC in patients with hypertension.

Fiber optic nanogold-linked immunosorbent assay for rapid detection of procalcitonin at femtomolar concentration level.

A rapid and ultrasensitive biosensing method based on fiber optic nanogold-linked immunosorbent assay is reported. The method employs an immobilized capture probe on the fiber core surface of an optical fiber and a detection probe conjugated to gold nanoparticles (AuNPs) in a solution. Introduction of a sample containing an analyte and the detection probe into a biosensor chip leads to the formation of a sandwich-like complex of capture probe-analyte-detection probe on the fiber core surface, through which nanoplasmonic absorption of the fiber optic evanescent wave occurs. The performance of this method has been evaluated by its application to the detection of procalcitonin (PCT), an important biomarker for sepsis. In this study, anti-PCT capture antibody is functionalized on an unclad segment of an optical fiber to yield a fiber sensor and anti-PCT detection antibody is conjugated to AuNPs to afford nanoplasmonic probes. The method provides a wide linear response range from 1 pg/mL to 100 ng/mL (5 orders) and an extremely low limit of detection of 95 fg/mL (7.3 fM) for PCT. In addition, the method shows a good correlation in determining PCT in blood plasma with the clinically validated electrochemiluninescent immunoassay. Furthermore, the method is quick (analysis time ≤15 min), requires low-cost instrumentation and sensor chips, and is also potentially applicable to the detection of many other biomarkers.

Bispectrum and Histogram Features for the Identification of Atrial Fibrillation Based on Electrocardiogram.

Atrial Fibrillation (AF) is probably the most common serious abnormal heart rhythm. It affects about 2% to 3% of the population in Europe and North America. In this study, we proposed an effective Atrial Fibrillation (AF) identification system based on RR interval (RRI) analysis. Two preprocessing methods were employed to remove the motion artifacts and ectopic beats. Three categories of RRI features, including base, bispectrum, and histogram features, were proposed to enhance the performance of the identifier. The roles of different feature categories were evaluated. The combination of the three categories of features were demonstrated to compensate with one another to construct an effective feature set for AF identification. When compared to other representative AF identifiers in the literature, the proposed method outperforms them with superior recognition rates by using much larger number of testing data.

Subband higher-order statistics and cross-correlation for heartbeat type recognition based on two-lead electrocardiogram.

Regular electrocardiogram beat classification system usually based on single lead ECG signal. This study designated to add a second lead of ECG signal to the system and apply higher-order statistics and inter-lead cross-correlation features to study the influence of the second lead to the recognition rates and noise-tolerance of the classifier. Discrete wavelet transformation is employed to decompose the ECG signals into different subband components and higher order statistics is recruited to characterize the ECG signals as an attempt to elevate the accuracy and noise-resistibility of heartbeat discrimination. A feed-forward back-propagation neural network (FFBNN) is employed as classifier. When compared with the system that uses only one lead, the second lead raises the recognition rate from 97.74% to 98.25%. We also study the ability of the two-lead system in resisting different levels of white Gaussian noise. More than 97.8% accuracy can be retained with the two-lead system even when the SNR decreases to 10 dB.

A tactile vision substitution system for the study of active sensing.

This paper presents a tactile vision substitution system (TVSS) for the study of active sensing. Two algorithms, namely image processing and trajectory tracking, were developed to enhance the capability of conventional TVSS. Image processing techniques were applied to reduce the artifacts and extract important features from the active camera and effectively converted the information into tactile stimuli with much lower resolution. A fixed camera was used to record the movement of the active camera. A trajectory tracking algorithm was developed to analyze the active sensing strategy of the TVSS users to explore the environment. The image processing subsystem showed advantageous improvement in extracting object's features for superior recognition. The trajectory tracking subsystem, on the other hand, enabled accurately locating the portion of the scene pointed by the active camera and providing profound information for the study of active sensing strategy applied by TVSS users.

A portable real-time ECG recognition system based on smartphone.

This paper proposed an smartphone-based real-time ECG monitoring and recognition system. The ECG signal was acquired by a MSP430FG4618 low-power microprocessor and was converted via a Bluetooth module for wireless transmission to a smartphone. A noise-tolerant ECG heartbeat recognition algorithm based on discrete wavelet transform and higher-order statistics was employed to identify different types of heartbeats. This system achieved a high accuracy of 98.34 % in identifying seven heartbeat types, which was demonstrated to outperform other studies in the literature. The heartbeat types were recognized in real-time; only 78 ms was required to identify a heartbeat. The portability, real-time processing, and high recognition rate of the system demonstrate the efficiency and effectiveness of the device as a practical computer-aided diagnosis (CAD) system.

Detection of myocardial ischemia episode using morphological features.

In this study, we propose to use morphological features that are easy to identify to differentiate myocardial ischemic beats from normal beats. In general, myocardial ischemia causes alterations in electrocardiographic (ECG) signal such as deviation in the ST segment. When the ST segment level deviates from a certain voltage, the beat would be diagnosing as myocardial ischemia. To emphasize on ST variations, the QRS complex of the ECG signal was first subtracted and replaced with a straight line. Five-level discrete wavelet transform (DWT) followed to decompose the waveform into subband components and the A5 subband, which is most sensitive to the changes in the ST segment, was reconstructed for the calculation of 12 morphological features. The support vector machine (SVM) and the 10-fold cross-validation method were employed to evaluate the performance of the method. The results show high values of 95.20%, 93.29%, and, 93.63% in sensitivity, specificity, and accuracy, respectively, that were demonstrated to outperform the other methods in the literature.

Bispectral analysis and genetic algorithm for congestive heart failure recognition based on heart rate variability.

This paper proposes a congestive heart failure (CHF) recognition method that includes features calculated from the bispectrum of heart rate variability (HRV) diagrams and a genetic algorithm (GA) for feature selection. The roles of the bispectrum-related features and the GA feature selector are investigated. Features calculated from the subband regions of the HRV bispectrum are added into a feature set containing only regular time-domain and frequency-domain features. A support vector machine (SVM) is employed as the classifier. A feature selector based on genetic algorithm proceeds to select the most effective features for the classifier. The results confirm the effectiveness of including bispectrum-related features for promoting the discrimination power of the classifier. When compared with the other two methods in the literature, the proposed method (without GA) outperforms both of them with a high accuracy of 96.38%. More than 3.14% surpluses in accuracies are observed. The application of GA as a feature selector further elevates the recognition accuracy from 96.38% to 98.79%. When compared to the Isler and Kuntalp's impressive results recently published in the literature that also uses GA for feature selection, the proposed method (with GA) outperforms them with more than 2.4% surpass in the recognition accuracy. These results confirm the significance of recruiting bispectrum-related features in a CHF classification system. Moreover, the application of GA as feature selector can further improve the performance of the classifier.

Conditional mutual information-based feature selection for congestive heart failure recognition using heart rate variability.

Feature selection plays an important role in pattern recognition systems. In this study, we explored the problem of selecting effective heart rate variability (HRV) features for recognizing congestive heart failure (CHF) based on mutual information (MI). The MI-based greedy feature selection approach proposed by Battiti was adopted in the study. The mutual information conditioned by the first-selected feature was used as a criterion for feature selection. The uniform distribution assumption was used to reduce the computational load. And, a logarithmic exponent weighting was added to model the relative importance of the MI with respect to the number of the already-selected features. The CHF recognition system contained a feature extractor that generated four categories, totally 50, features from the input HRV sequences. The proposed feature selector, termed UCMIFS, proceeded to select the most effective features for the succeeding support vector machine (SVM) classifier. Prior to feature selection, the 50 features produced a high accuracy of 96.38%, which confirmed the representativeness of the original feature set. The performance of the UCMIFS selector was demonstrated to be superior to the other MI-based feature selectors including MIFS-U, CMIFS, and mRMR. When compared to the other outstanding selectors published in the literature, the proposed UCMIFS outperformed them with as high as 97.59% accuracy in recognizing CHF using only 15 features. The results demonstrated the advantage of using the recruited features in characterizing HRV sequences for CHF recognition. The UCMIFS selector further improved the efficiency of the recognition system with substantially lowered feature dimensions and elevated recognition rate.

Selection of effective features for ECG beat recognition based on nonlinear correlations.

OBJECTIVE: The objective of this study is to develop feature selectors based on nonlinear correlations in order to select the most effective and least redundant features from an ECG beat classification system based on higher order statistics of subband components and a feed-forward back-propagation neural network, denoted as HOS-DWT-FFBNN. METHODS AND MATERIALS: Three correlation-based filters (NCBFs) are proposed. Two of them, NCBF1 and NCBF2, apply feature-feature correlation to remove redundant features prior to the feature selection process based on feature-class correlation. The other, SUFCO, skips the redundancy reduction process and selects features based only on feature-class correlation. The performance of these filters is compared to another commonly used nonlinear feature selection method, Relief-F. The discriminality and redundancy of the retained features are evaluated quantitatively. The performance of the most effective NCBF is compared with that of the linear correlation-based filter (LCBF) and other representative heartbeat classifiers in the literature. RESULTS: The results demonstrate that the two NCBFs based on both feature-feature and feature-class correlation methods, i.e. NCBF1 and NCBF2, outperform the other two methods, i.e. SUFCO and Relief-F. An accuracy of as high as 96.34% can be attained with as few as eight features. When tested with statistical methods, the retained features selected by the NCBF1/NCBF2 approach are demonstrated to be more discriminative and less redundant when compared with those features selected by other methods. When compared with LCBF and other heartbeat classifiers in the literature, the proposed NCBF1/NCBF2 approach in conjunction with the HOS-DWT-FFBNN structure outperform them with improved performance that allows discrimination of more beat types and fewer feature dimensions. CONCLUSION: This study demonstrates the effectiveness and superiority of the proposed approach for ECG beat recognition.

Multiscale sample entropy based on discrete wavelet transform for clinical heart rate variability recognition.

Traditional multiscale method uses coarse grained average (CGA) to evaluate sample entropy (SE) parameters in different scales for signal characterization. In this study, we propose to use discrete wavelet transform (DWT) to decompose hear rate variability signals into multiscale sequences for the calculation of SE features for the recognition of congestive heart failure (CHF) and atrial fibrillation (AF) from normal sinus rhythm (NSR). The support vector machine (SVM) is used as the classifier and the capability of the features are justified with leave-one-out cross-validation method. The results demonstrate that the system using multiscale SE features calculated from both CGA and DWT with five dyadic scales outperforms that based on tradition multiscale method using CGA and 20 scales. Compared to the 5-scale CGA method, the proposed 5-scale DWT method achieved 6.7% and 0.77% increases in the recognition rates for CHF and AF, respectively, and resulted in an 8.35% raise in the overall recognition accuracy.

Improving discriminality in heart rate variability analysis using simple artifact and trend removal preprocessors.

Heart Rate variability (HRV) is important in characterizing heart functions. However, artifacts and trends are regularly observed to contaminate the HRV sequences. This study proposes a simple and effective preprocessor for the removal of artifacts and trend in the HRV sequences. A thresholding filter is applied to remove artifacts to maintain the HRV sequences in a reasonable range. A wavelet filter proceeds to remove the ultra and very low frequency components determined as trends. As a consequence, more reliable low frequency (LF) and high frequency (HF) components can be calculated, which are believed to be close-related to the autonomic nervous system (ANS) regulation of the heart. The result demonstrates that features calculated from the power spectral density of the preprocessed HRV are more separable in feature space when compared with that from the original HRV. A simple KNN classifier is employed to justify the effects of this preprocessor in differentiating congestive heart failure (CHF) from the normal sinus rhythms (NSR). Using five features calculated from LF and HF, the performance of the KNN classifier shows significant improvement after applying the preprocessors. When compared with the other studies published in the literature, the proposed method outperforms them in CHF recognition with a much simpler scheme.

Detection of seizures in EEG using subband nonlinear parameters and genetic algorithm.

Detection of seizures in EEG can be challenging because of myogenic artifacts and might be time-consuming. In this study, we propose a method using subband nonlinear parameters and genetic algorithm for automatic seizure detection in EEG. In the experiment, the discrete wavelet transform was used to decompose EEG into five subband components. Nonlinear parameters were extracted and employed as the features to train the support vector machine with linear kernel function (SVML) and radial basis function kernel function (SVMRBF) classifiers. A genetic algorithm (GA) was used for selecting the effective feature subset. The seizure detection sensitivities of the SVML and the SVMRBF with GA are 90.8% and 94.0%, respectively. The sensitivity of SVMRBF increases to 95.8% by using GA for weight adjustment. Moreover, the proposed method is capable of discriminating the interictal EEG of epileptic subjects from the normal EEG, which is considered difficult, yet crucial, in clinical services.

Improving liquid chromatography-tandem mass spectrometry determinations by modifying noise frequency spectrum between two consecutive wavelet-based low-pass filtering procedures.

This paper employs one chemometric technique to modify the noise spectrum of liquid chromatography-tandem mass spectrometry (LC-MS/MS) chromatogram between two consecutive wavelet-based low-pass filter procedures to improve the peak signal-to-noise (S/N) ratio enhancement. Although similar techniques of using other sets of low-pass procedures such as matched filters have been published, the procedures developed in this work are able to avoid peak broadening disadvantages inherent in matched filters. In addition, unlike Fourier transform-based low-pass filters, wavelet-based filters efficiently reject noises in the chromatograms directly in the time domain without distorting the original signals. In this work, the low-pass filtering procedures sequentially convolve the original chromatograms against each set of low pass filters to result in approximation coefficients, representing the low-frequency wavelets, of the first five resolution levels. The tedious trials of setting threshold values to properly shrink each wavelet are therefore no longer required. This noise modification technique is to multiply one wavelet-based low-pass filtered LC-MS/MS chromatogram with another artificial chromatogram added with thermal noises prior to the other wavelet-based low-pass filter. Because low-pass filter cannot eliminate frequency components below its cut-off frequency, more efficient peak S/N ratio improvement cannot be accomplished using consecutive low-pass filter procedures to process LC-MS/MS chromatograms. In contrast, when the low-pass filtered LC-MS/MS chromatogram is conditioned with the multiplication alteration prior to the other low-pass filter, much better ratio improvement is achieved. The noise frequency spectrum of low-pass filtered chromatogram, which originally contains frequency components below the filter cut-off frequency, is altered to span a broader range with multiplication operation. When the frequency range of this modified noise spectrum shifts toward the high frequency regimes, the other low-pass filter is able to provide better filtering efficiency to obtain higher peak S/N ratios. Real LC-MS/MS chromatograms, of which typically less than 6-fold peak S/N ratio improvement achieved with two consecutive wavelet-based low-pass filters remains the same S/N ratio improvement using one-step wavelet-based low-pass filter, are improved to accomplish much better ratio enhancement 25-folds to 40-folds typically when the noise frequency spectrum is modified between two low-pass filters. The linear standard curves using the filtered LC-MS/MS signals are validated. The filtered LC-MS/MS signals are also reproducible. The more accurate determinations of very low concentration samples (S/N ratio about 7-9) are obtained using the filtered signals than the determinations using the original signals.

Noise-tolerant electrocardiogram beat classification based on higher order statistics of subband components.

OBJECTIVE: This paper presents a noise-tolerant electrocardiogram (ECG) beat classification method based on higher order statistics (HOS) of subband components. METHODS AND MATERIAL: Five levels of discrete wavelet transform (DWT) were applied to decompose the signal into six subband components. Higher order statistics proceeded to calculate four sets of HOS features from the three midband components, which together with three RR interval-related features constructed the primary feature set. A feature selection algorithm based on correlation coefficient and Fisher discriminality was then exploited to eliminate redundant features from the primary feature set. A feedforward backpropagation neural network (FFBNN) was employed as the classifier. Two sample selection strategies and four categories of noise artifacts were utilized to justify the capacity of the method. RESULTS: More than 97.5% discrimination rate was achieved, no matter which of the two sampling selection strategies was used. By using the feature selection method, the feature dimension can be readily reduced from 30 to 18 with negligible decrease in accuracy. Compared with other method in the literature, the proposed method improves the sensitivities of most beat types, resulting in an elevated average accuracy. The proposed method is tolerant to environmental noises; as high as 91% accuracies were retained even when contaminated with serious noises, 10 dB signal-to-noise ration (SNR), of different kinds. CONCLUSION: The results demonstrate the effectiveness and noise-tolerant capacities of the proposed method in ECG beat classification.

Characterization of the major histopathological components of thyroid nodules using sonographic textural features for clinical diagnosis and management.

In this study, the characteristic sonographic textural feature that represents the major histopathologic components of the thyroid nodules was objectively quantified to facilitate clinical diagnosis and management. A total of 157 regions-of-interest thyroid ultrasound image was recruited in the study. The sonographic system used was the GE LOGIQ 700), (General Electric Healthcare, Chalfant St. Giles, UK). The parameters affecting image acquisition were kept in the same condition for all lesions. Commonly used texture analysis methods were applied to characterize thyroid ultrasound images. Image features were classified according to the corresponding pathologic findings. To estimate their relevance and performance to classification, ReliefF was used as a feature selector. Among the various textural features, the sum average value derived from co-occurrence matrix can well reflect echogenicity and can effectively differentiate between follicles and fibrosis base thyroid nodules. Fibrosis shows lowest echogenicity and lowest difference sum average value. Enlarged follicles show highest echogenicity and difference sum average values. Papillary cancer or follicular tumors show the difference sum average values and echogenicity between. The rule of thumb for the echogenicity is that the more follicles are mixed in, the higher the echo of the follicular tumor and papillary cancer will be and vice versa for fibrosis mixed. Areas with intermediate and lower echo should address the possibility of follicular or papillary neoplasm mixed with either follicles or fibrosis. These areas provide more cellular information for ultrasound guided aspiration

Quantitative correlation between sonographic textural feature and histopathological components for breast cancer: preliminary results.

In this study, the sonographic texture and the histopathological features of breast cancer were objectively characterized. Textural dissimilarity is demonstrated to correlate well with the corresponding histopathological components. The normalized percentage of both fibrosis area and cellular area has highly linear correlation with the textural feature of dissimilarity. The correlation coefficients are -.880 and .857, respectively. The cancerous region with increased fibrous tissues shows low textural dissimilarity and has a strong tendency of negative correlation, whereas the cancerous region with increased cellularity exhibits high textural dissimilarity and a good positive correlation. These results have not been reported so far, and they can be used to predict cellular and fibrotic portions of breast cancer for biopsy or surgery planning, disease progression monitoring, and therapeutic effect evaluation. The proposed image analysis method may also be extended to similar characterization of cancerous tissue in other applications.

Tailoring noise frequency spectrum between two consecutive second derivative filtering procedures to improve liquid chromatography-mass spectrometry determinations.

This paper reports a simple chemometric technique to alter the noise spectrum of a liquid chromatography-mass spectrometry (LC-MS) chromatogram between two consecutive second-derivative filter procedures to improve the peak signal-to-noise (S/N) ratio enhancement. This technique is to multiply one second-derivative filtered LC-MS chromatogram with another artificial chromatogram added with thermal noises prior to the other second-derivative filter. Because the second-derivative filter cannot eliminate frequency components within its own filter bandwidth, more efficient peak S/N ratio improvement cannot be accomplished using consecutive second-derivative filter procedures to process LC-MS chromatograms. In contrast, when the second-derivative filtered LC-MS chromatogram is conditioned with the multiplication alteration prior to the other second-derivative filter, much better ratio improvement is achieved. The noise frequency spectrum of the second-derivative filtered chromatogram, which originally contains frequency components within the filter bandwidth, is altered to span a broader range with multiplication operation. When the frequency range of this modified noise spectrum shifts toward the other regimes, the other second-derivative filter, working as a band-pass filter, is able to provide better filtering efficiency to obtain higher peak S/N ratios. Real LC-MS chromatograms, of which 5-fold peak S/N ratio improvement achieved with two consecutive second-derivative filters remains the same S/N ratio improvement using a one-step second-derivative filter, are improved to accomplish much better ratio enhancement, approximately 25-fold or higher when the noise frequency spectrum is modified between two matched filters. The linear standard curve using the filtered LC-MS signals is validated. The filtered LC-MS signals are also more reproducible. The more accurate determinations of very low-concentration samples (S/N ratio about 5-7) are obtained via standard addition procedures using the filtered signals rather than the determinations using the original signals.