Decision and feature level fusion of deep features extracted from public COVID-19 data-sets.

The Coronavirus disease (COVID-19), which is an infectious pulmonary disorder, has affected millions of people and has been declared as a global pandemic by the WHO. Due to highly contagious nature of COVID-19 and its high possibility of causing severe conditions in the patients, the development of rapid and accurate diagnostic tools have gained importance. The real-time reverse transcription-polymerize chain reaction (RT-PCR) is used to detect the presence of Coronavirus RNA by using the mucus and saliva mixture samples taken by the nasopharyngeal swab technique. But, RT-PCR suffers from having low-sensitivity especially in the early stage. Therefore, the usage of chest radiography has been increasing in the early diagnosis of COVID-19 due to its fast imaging speed, significantly low cost and low dosage exposure of radiation. In our study, a computer-aided diagnosis system for X-ray images based on convolutional neural networks (CNNs) and ensemble learning idea, which can be used by radiologists as a supporting tool in COVID-19 detection, has been proposed. Deep feature sets extracted by using seven CNN architectures were concatenated for feature level fusion and fed to multiple classifiers in terms of decision level fusion idea with the aim of discriminating COVID-19, pneumonia and no-finding classes. In the decision level fusion idea, a majority voting scheme was applied to the resultant decisions of classifiers. The obtained accuracy values and confusion matrix based evaluation criteria were presented for three progressively created data-sets. The aspects of the proposed method that are superior to existing COVID-19 detection studies have been discussed and the fusion performance of proposed approach was validated visually by using Class Activation Mapping technique. The experimental results show that the proposed approach has attained high COVID-19 detection performance that was proven by its comparable accuracy and superior precision/recall values with the existing studies.

Tunable Q-factor wavelet transform based lung signal decomposition and statistical feature extraction for effective lung disease classification.

The auscultation is a non-invasive and cost-effective method used for the diagnosis of lung diseases, which are one of the leading causes of death worldwide. However, the efficacy of the auscultation suffers from the limitations of the analog stethoscopes and the subjective nature of human interpretation. To overcome these limitations, the accurate diagnosis of these diseases by employing the computer based automated algorithms applied to the digitized lung sounds has been studied for the last decades. This study proposes a novel approach that uses a Tunable Q-factor Wavelet Transform (TQWT) based statistical feature extraction followed by individual and ensemble learning model training with the aim of lung disease classification. During the learning stage various machine learning algorithms are utilized as the individual learners as well as the hard and soft voting fusion approaches are employed for performance enhancement with the aid of the predictions of individual models. For an objective evaluation of the proposed approach, the study was structured into two main tasks that were investigated in detail by using several sub-tasks to comparison with state-of-the-art studies. Among the sub-tasks which investigates patient-based classification, the highest accuracy obtained for the binary classification was achieved as 97.63% (healthy vs. non-healthy), while accuracy values up to 66.32% for three-class classification (obstructive-related, restrictive-related, and healthy), and 53.42% for five-class classification (asthma, chronic obstructive pulmonary disease, interstitial lung disease, pulmonary infection, and healthy) were obtained. Regarding the other sub-task, which investigates sample-based classification, the proposed approach was superior to almost all previous findings. The proposed method underscores the potential of TQWT based signal decomposition that leverages the power of its adaptive time-frequency resolution property satisfied by Q-factor adjustability. The obtained results are very promising and the proposed approach paves the way for more accurate and automated digital auscultation techniques in clinical settings.

Resonance based separation and energy based classification of lung sounds using tunable wavelet transform.

BACKGROUND AND OBJECTIVE: The locations and occurrence pattern of adventitious sounds in the respiratory cycle have critical diagnostic information. In a lung sound sample, the crackles and wheezes may exist individually or they may coexist in a successive/overlapping manner superimposed onto the breath noise. The performance of the linear time-frequency representation based signal decomposition methods has been limited in the crackle/wheeze separation problem due to the common signal components that may arise in both time and frequency domain. However, the proposed resonance based decomposition can be used to isolate crackles and wheezes which behave oppositely in time domain even if they share common frequency bands. METHODS: In the proposed study, crackle and/or wheeze containing synthetic and recorded lung-sound signals were decomposed by using the resonance information which is produced by joint application of the Tunable Q-factor Wavelet Transform and Morphological Component Analysis. The crackle localization and signal reconstruction performance of the proposed approach was compared with the previously suggested Independent Component Analysis and Empirical Mode Decomposition methods in a quantitative and qualitative manner. Additionally, the decomposition ability of the proposed approach was also used to discriminate crackle and wheeze waveforms in an unsupervised way by employing signal energy. RESULTS: Results have shown that the proposed approach has significant superiority over its competitors in terms of the crackle localization and signal reconstruction ability. Moreover, the calculated energy values have revealed that the transient crackles and rhythmic wheezes can be successfully decomposed into low and high resonance channels by preserving the discriminative information. CONCLUSIONS: It is concluded that previous works suffer from deforming the waveform of the crackles whose time domain parameters are vital in computerized diagnostic classification systems. Therefore, a method should provide automatic and simultaneous decomposition ability, with smaller root mean square error and higher accuracy as demonstrated by the proposed approach.

Automated sperm morphology analysis approach using a directional masking technique.

Sperm Morphology is the key step in the assessment of sperm quality. Due to the effect of misleading human factors in manual assessments, computer-based techniques should be employed in the analysis. In this study, a computation framework including multi-stage cascade connected preprocessing techniques, region based descriptor features, and non-linear kernel SVM based learning is proposed for the classification of any stained sperm images for the assessment of the morphology. The proposed framework was evaluated on two sperm morphology datasets: the Human Sperm Head Morphology dataset (HuSHeM) and Sperm Morphology Image Data Set (SMIDS). The results indicate that cascading the preprocessing techniques used in the proposed framework, such as wavelet based local adaptive de-noising, modified overlapping group shrinkage, image gradient, and automatic directional masking, increased the classification accuracy by 10% and 5% for the HuSHeM and SMIDS, respectively. The proposed framework results in better overall accuracy than most state-of-the-art methods, while having significant advantages, such as eliminating the exhaustive manual orientation and cropping operations of the competitors with reasonable rates of consumption of time and source.

A fully automated hybrid human sperm detection and classification system based on mobile-net and the performance comparison with conventional methods.

Sperm morphology, as an indicator of fertility, is a critical tool in semen analysis. In this study, a smartphone-based hybrid system that fully automates the sperm morphological analysis is introduced with the aim of eliminating unwanted human factors. Proposed hybrid system consists of two progressive steps: automatic segmentation of possible sperm shapes and classification of normal/ab-normal sperms. In the segmentation step, clustering techniques with/without group sparsity approach were tested to extract region of interests from the images. Subsequently, a novel publicly available morphological sperm image data set, whose labels were identified by experts as non-sperm, normal and abnormal sperm, was created as the ground truths of classification step. In the classification step, conventional and ensemble machine learning methods were applied to domain-specific features that were extracted by using wavelet transform and descriptors. Additionally, as an alternative to conventional features, three deep neural network architectures, which can extract high-level features from raw images after using statistical learning, were employed to increase the proposed method's performance. The results show that, for the conventional features, the highest classification accuracies were achieved as 80.5% and 83.8% by using the wavelet- and descriptor-based features that were fed to the Support Vector Machines respectively. On the other hand, the Mobile-Net, which is a very convenient network for smartphones, achieved 87% accuracy. In the light of obtained results, it is seen that a fully automatic hybrid system, which uses the group sparsity to enhance segmentation performance and the Mobile-Net to obtain high-level robust features, can be an effective mobile solution for the sperm morphology analysis problem. A fully automated hybrid human sperm detection and classification system based on mobile-net.

Estimation of Parkinson's disease severity using speech features and extreme gradient boosting.

In recent years, there is an increasing interest in building e-health systems. The systems built to deliver the health services with the use of internet and communication technologies aim to reduce the costs arising from outpatient visits of patients. Some of the related recent studies propose machine learning-based telediagnosis and telemonitoring systems for Parkinson's disease (PD). Motivated from the studies showing the potential of speech disorders in PD telemonitoring systems, in this study, we aim to estimate the severity of PD from voice recordings of the patients using motor Unified Parkinson's Disease Rating Scale (UPDRS) as the evaluation metric. For this purpose, we apply various speech processing algorithms to the voice signals of the patients and then use these features as input to a two-stage estimation model. The first step is to apply a wrapper-based feature selection algorithm, called Boruta, and select the most informative speech features. The second step is to feed the selected set of features to a decision tree-based boosting algorithm, extreme gradient boosting, which has been recently applied successfully in many machine learning tasks due to its generalization ability and speed. The feature selection analysis showed that the vibration pattern of the vocal fold is an important indicator of PD severity. Besides, we also investigate the effectiveness of using age and years passed since diagnosis as covariates together with speech features. The lowest mean absolute error with 3.87 was obtained by combining these covariates and speech features with prediction level fusion. Graphical Abstract Framework for the proposed UPDRS estimation model.

Wheeze type classification using non-dyadic wavelet transform based optimal energy ratio technique.

BACKGROUND AND OBJECTIVE: Wheezes in pulmonary sounds are anomalies which are often associated with obstructive type of lung diseases. The previous works on wheeze-type classification focused mainly on using fixed time-frequency/scale resolution based on Fourier and wavelet transforms. The main contribution of the proposed method, in which the time-scale resolution can be tuned according to the signal of interest, is to discriminate monophonic and polyphonic wheezes with higher accuracy than previously suggested time and time-frequency/scale based methods. METHODS: An optimal Rational Dilation Wavelet Transform (RADWT) based peak energy ratio (PER) parameter selection method is proposed to discriminate wheeze types. Previously suggested Quartile Frequency Ratios, Mean Crossing Irregularity, Multiple Signal Classification, Mel-frequency Cepstrum and Dyadic Discrete Wavelet Transform approaches are also applied and the superiority of the proposed method is demonstrated in leave-one-out (LOO) and leave-one-subject-out (LOSO) cross validation schemes with support vector machine (SVM), k nearest neighbor (k-NN) and extreme learning machine (ELM) classifiers. RESULTS: The results show that the proposed RADWT based method outperforms the state-of-the-art time, frequency, time-frequency and time-scale domain approaches for all classifiers in both LOO and LOSO cross validation settings. The highest accuracy values are obtained as 86% and 82.9% in LOO and LOSO respectively when the proposed PER features are fed into SVM. CONCLUSIONS: It is concluded that time and frequency domain characteristics of wheezes are not steady and hence, tunable time-scale representations are more successful in discriminating polyphonic and monophonic wheezes when compared with conventional fixed resolution representations.

An open access database for the evaluation of respiratory sound classification algorithms.

OBJECTIVE: Over the last few decades, there has been significant interest in the automatic analysis of respiratory sounds. However, currently there are no publicly available large databases with which new algorithms can be evaluated and compared. Further developments in the field are dependent on the creation of such databases. APPROACH: This paper describes a public respiratory sound database, which was compiled for an international competition, the first scientific challenge of the IFMBE's International Conference on Biomedical and Health Informatics. The database includes 920 recordings acquired from 126 participants and two sets of annotations. One set contains 6898 annotated respiratory cycles, some including crackles, wheezes, or a combination of both, and some with no adventitious respiratory sounds. In the other set, precise locations of 10 775 events of crackles and wheezes were annotated. MAIN RESULTS: The best system that participated in the challenge achieved an average score of 52.5% with the respiratory cycle annotations and an average score of 91.2% with the event annotations. SIGNIFICANCE: The creation and public release of this database will be useful to the research community and could bring attention to the respiratory sound classification problem.

Performance comparison of wavelet based denoising methods on discontinuous adventitious lung sounds.

Crackles and their time-domain characteristics provide important clues about different lung diseases. In this paper, we aim to de-noise synthetically produced crackles under various noise levels while preserving their information bearing parts which significantly affect crackle parameters. Classical wavelet based de-noising algorithms are deteriorated by sharp-sudden noise changes and produce Gibbs like fluctuations. On the other hand, total variation based algorithms, which are capable of alleviating the drawbacks of the classical wavelet based algorithms, are failed when dealing with piecewise-smooth signals like crackles and generate unwanted flat regions on the de-noised signals. Proposed wavelet total variation based de-noising is succeed in removing undesired artefacts originating from both classical wavelet and total variation de-noising. The proposed method is compared with classical wavelet based de-noising methods in terms of root mean square error under various white Gaussian noise levels (0 - 20 dB SNR). Moreover, in order to emphasize the de-noising ability of the methods, without deforming crackle waveform, time and frequency domain representation of a noisy and de-noised crackle is validated visually.

Analyzing the effectiveness of vocal features in early telediagnosis of Parkinson's disease.

The recently proposed Parkinson's Disease (PD) telediagnosis systems based on detecting dysphonia achieve very high classification rates in discriminating healthy subjects from PD patients. However, in these studies the data used to construct the classification model contain the speech recordings of both early and late PD patients with different severities of speech impairments resulting in unrealistic results. In a more realistic scenario, an early telediagnosis system is expected to be used in suspicious cases by healthy subjects or early PD patients with mild speech impairment. In this paper, considering the critical importance of early diagnosis in the treatment of the disease, we evaluate the ability of vocal features in early telediagnosis of Parkinson's Disease (PD) using machine learning techniques with a two-step approach. In the first step, using only patient data, we aim to determine the patient group with relatively greater severity of speech impairments using Unified Parkinson's Disease Rating Scale (UPDRS) score as an index of disease progression. For this purpose, we use three supervised and two unsupervised learning techniques. In the second step, we exclude the samples of this group of patients from the dataset, create a new dataset consisting of the samples of PD patients having less severity of speech impairments and healthy subjects, and use three classifiers with various settings to address this binary classification problem. In this classification problem, the highest accuracy of 96.4% and Matthew's Correlation Coefficient of 0.77 is obtained using support vector machines with third-degree polynomial kernel showing that vocal features can be used to build a decision support system for early telediagnosis of PD.

Analysis of embolic signals with directional dual tree rational dilation wavelet transform.

The dyadic discrete wavelet transform (dyadic-DWT), which is based on fixed integer sampling factor, has been used before for processing piecewise smooth biomedical signals. However, the dyadic-DWT has poor frequency resolution due to the low-oscillatory nature of its wavelet bases and therefore, it is less effective in processing embolic signals (ESs). To process ESs more effectively, a wavelet transform having better frequency resolution than the dyadic-DWT is needed. Therefore, in this study two ESs, containing micro-emboli and artifact waveforms, are analyzed with the Directional Dual Tree Rational-Dilation Wavelet Transform (DDT-RADWT). The DDT-RADWT, which can be directly applied to quadrature signals, is based on rational dilation factors and has adjustable frequency resolution. The analyses are done for both low and high Q-factors. It is proved that, when high Q-factor filters are employed in the DDT-RADWT, clearer representations of ESs can be attained in decomposed sub-bands and artifacts can be successfully separated.

Directional dual-tree complex wavelet packet transforms for processing quadrature signals.

Quadrature signals containing in-phase and quadrature-phase components are used in many signal processing applications in every field of science and engineering. Specifically, Doppler ultrasound systems used to evaluate cardiovascular disorders noninvasively also result in quadrature format signals. In order to obtain directional blood flow information, the quadrature outputs have to be preprocessed using methods such as asymmetrical and symmetrical phasing filter techniques. These resultant directional signals can be employed in order to detect asymptomatic embolic signals caused by small emboli, which are indicators of a possible future stroke, in the cerebral circulation. Various transform-based methods such as Fourier and wavelet were frequently used in processing embolic signals. However, most of the times, the Fourier and discrete wavelet transforms are not appropriate for the analysis of embolic signals due to their non-stationary time-frequency behavior. Alternatively, discrete wavelet packet transform can perform an adaptive decomposition of the time-frequency axis. In this study, directional discrete wavelet packet transforms, which have the ability to map directional information while processing quadrature signals and have less computational complexity than the existing wavelet packet-based methods, are introduced. The performances of proposed methods are examined in detail by using single-frequency, synthetic narrow-band, and embolic quadrature signals.

Resonance based respiratory sound decomposition aiming at localization of crackles in noisy measurements.

In this work, resonance based decomposition of lung sounds that aims to separate wheeze, crackle and vesicular sounds into three individual channels while automatically localizing crackles for both synthetic and real data is presented. Previous works focus on stationary-non stationary discrimination to separate crackles and vesicular sounds disregarding wheezes which are stationary as compared to crackles. However, wheeze sounds include important cues about the underlying pathology. Using two different threshold methods and synthetic sound generation scenarios in the presence of wheezes, resonance based decomposition performs 89.5 % crackle localization recall rate for white Gaussian noise and 98.6 % crackle localization recall rate for healthy vesicular sound treated as noise at low signal-to-noise ratios. Besides, an adaptive threshold determination which is independent from the channel at which it will be applied is used and is found to be robust to noise.

A lung sound classification system based on the rational dilation wavelet transform.

In this work, a wavelet based classification system that aims to discriminate crackle, normal and wheeze lung sounds is presented. While the previous works related with this problem use constant low Q-factor wavelets, which have limited frequency resolution and can not cope with oscillatory signals, in the proposed system, the Rational Dilation Wavelet Transform, whose Q-factors can be tuned, is employed. Proposed system yields an accuracy of 95 % for crackle, 97 % for wheeze, 93.50 % for normal and 95.17 % for total sound signal types using energy feature subset and proposed approach is superior to conventional low Q-factor wavelet analysis.

Code generator for implementing dual tree complex wavelet transform on reconfigurable architectures for mobile applications.

The authors aimed to develop an application for producing different architectures to implement dual tree complex wavelet transform (DTCWT) having near shift-invariance property. To obtain a low-cost and portable solution for implementing the DTCWT in multi-channel real-time applications, various embedded-system approaches are realised. For comparison, the DTCWT was implemented in C language on a personal computer and on a PIC microcontroller. However, in the former approach portability and in the latter desired speed performance properties cannot be achieved. Hence, implementation of the DTCWT on a reconfigurable platform such as field programmable gate array, which provides portable, low-cost, low-power, and high-performance computing, is considered as the most feasible solution. At first, they used the system generator DSP design tool of Xilinx for algorithm design. However, the design implemented by using such tools is not optimised in terms of area and power. To overcome all these drawbacks mentioned above, they implemented the DTCWT algorithm by using Verilog Hardware Description Language, which has its own difficulties. To overcome these difficulties, simplify the usage of proposed algorithms and the adaptation procedures, a code generator program that can produce different architectures is proposed.

Field programmable gate arrays implementation of Dual Tree Complex Wavelet Transform.

Due to the inherent time-varying characteristics of physiological systems, most biomedical signals (BSs) are expected to have non-stationary character. Therefore, any appropriate analysis method for dealing with BSs should exhibit adjustable time-frequency (TF) resolution. The wavelet transform (WT) provides a TF representation of signals, which has good frequency resolution at low frequencies and good time resolution at high frequencies, resulting in an optimized TF resolution. Discrete wavelet transform (DWT), which is used in various medical signal processing applications such as denoising and feature extraction, is a fast and discretized algorithm for classical WT. However, the DWT has some very important drawbacks such as aliasing, lack of directionality, and shift-variance. To overcome these drawbacks, a new improved discrete transform named as Dual Tree Complex Wavelet Transform (DTCWT) can be used. Nowadays, with the improvements in embedded system technology, portable real-time medical devices are frequently used for rapid diagnosis in patients. In this study, in order to implement DTCWT algorithm in FPGAs, which can be used as real-time feature extraction or denoising operator for biomedical signals, a novel hardware architecture is proposed. In proposed architecture, DTCWT is implemented with only one adder and one multiplier. Additionally, considering the multi-channel outputs of biomedical data acquisition systems, this architecture is capable of running N channels in parallel.

Denoising performance of modified dual-tree complex wavelet transform for processing quadrature embolic Doppler signals.

Quadrature signals are dual-channel signals obtained from the systems employing quadrature demodulation. Embolic Doppler ultrasound signals obtained from stroke-prone patients by using Doppler ultrasound systems are quadrature signals caused by emboli, which are particles bigger than red blood cells within circulatory system. Detection of emboli is an important step in diagnosing stroke. Most widely used parameter in detection of emboli is embolic signal-to-background signal ratio. Therefore, in order to increase this ratio, denoising techniques are employed in detection systems. Discrete wavelet transform has been used for denoising of embolic signals, but it lacks shift invariance property. Instead, dual-tree complex wavelet transform having near-shift invariance property can be used. However, it is computationally expensive as two wavelet trees are required. Recently proposed modified dual-tree complex wavelet transform, which reduces the computational complexity, can also be used. In this study, the denoising performance of this method is extensively evaluated and compared with the others by using simulated and real quadrature signals. The quantitative results demonstrated that the modified dual-tree-complex-wavelet-transform-based denoising outperforms the conventional discrete wavelet transform with the same level of computational complexity and exhibits almost equal performance to the dual-tree complex wavelet transform with almost half computational cost.

Symmetrical directional dual-tree complex wavelet packet transform.

In this study, a symmetrical directional complex discrete wavelet packet transform, which can be applied directly to the quadrature format signals and has the ability of mapping directional information during decomposition stage, is proposed. With the proposed symmetrical directional complex wavelet packet transform, traditional symmetrical phasing filter technique, which is used for quadrature signal to directional signal conversion, is eliminated and the computational complexity of whole process is reduced. The performance of proposed method is examined in detail using real quadrature embolic signals.

Directional dual-tree rational-dilation complex wavelet transform.

Dyadic discrete wavelet transform (DWT) has been used successfully in processing signals having non-oscillatory transient behaviour. However, due to the low Q-factor property of their wavelet atoms, the dyadic DWT is less effective in processing oscillatory signals such as embolic signals (ESs). ESs are extracted from quadrature Doppler signals, which are the output of Doppler ultrasound systems. In order to process ESs, firstly, a pre-processing operation known as phase filtering for obtaining directional signals from quadrature Doppler signals must be employed. Only then, wavelet based methods can be applied to these directional signals for further analysis. In this study, a directional dual-tree rational-dilation complex wavelet transform, which can be applied directly to quadrature signals and has the ability of extracting directional information during analysis, is introduced.

Directional dual-tree complex wavelet packet transform.

Doppler ultrasound systems, which are widely used in cardiovascular disorders detection, have quadrature format outputs. Various types of algorithms were described in literature to process quadrature Doppler signals (QDS), such as phasing filter technique (PFT), fast Fourier transform method, frequency domain Hilbert transform method and complex continuous wavelet transform. However for the discrete wavelet transform (DWT) case, which becomes a common method for processing QDSs, there was not a direct method to recover flow direction from quadrature signals. Traditionally, to process QDSs with DWT, firstly directional signals have to be extracted and later two DWTs must be applied. Although there exists a complex DWT algorithm called dual tree complex discrete wavelet transform (DTCWT), it does not provide directional signal decoding during analysis because of the unwanted energy leaks into its negative frequency bands. Modified DTCWT, which is a combination of PFT and DTCWT, has the capability of extracting directional information while decomposing QDSs into different frequency bands, but it uses an additional Hilbert transform filter and it increases the computational complexity of whole transform. Discrete wavelet packet transform (DWPT), which is a generalization of the ordinary DWT allowing subband analysis without the constraint of dyadic decomposition, can perform an adaptive decomposition of the frequency axis. In this study, a novel complex DWPT, which maps directional information while processing QDSs, is proposed. The success of proposed method will be measured by using simulated quadrature signals.