Background and purpose Auscultatory sounds acquired using a stethoscope can offer clinical clues to the presence of cardiorespiratory diseases. In this work, we aimed to develop an accurate and lightweight model for disease detection using lung sounds. Method Our model comprises: (1) signal decomposition using a multilevel bidirectional wavelet transformation; (2) multilevel feature generation using a novel lattice-based graphene pattern to create minimum- and maximum directed graphs to extract textural features; (3) feature selection using iterative neighborhood component analysis; (4) classification using a standard shallow k-nearest neighbor function. We tested the model on a public 336-subject eight-class lung sound dataset. Model performance was reported for eight- and three-class diagnostic classification. Results Our model achieved accuracy rates exceeding 94 % for all classification tasks. The maximum distance path through the graphene pattern consistently outperformed the minimum distance path, indicating that significant amplitude transitions in respiratory sounds contain more discriminative information than regions of relative uniformity. Elements of the input signal and wavelet decomposition bands that contributed most to the selected feature vector were visualized, which enhanced model explainability and revealed that low-pass filtered wavelet coefficients, particularly the L3 band, were most informative for classification. Conclusion Our handcrafted computationally lightweight model yielded accurate and explainable results. These attributes facilitate potential integration into digital stethoscopes for point-of-care screening of respiratory diseases.

Abstract Examining peripheral blood smears is valuable in clinical settings, yet manual identification of blood cells proves time-consuming. To address this, an automated blood cell image classification system is crucial. Our objective is to develop a precise automated model for detecting various blood cell types, leveraging a novel deep learning architecture. We harnessed a publicly available dataset of 17,092 blood cell images categorized into eight classes. Our innovation lies in ConcatNeXt, a new convolutional neural network. In the spirit of Geoffrey Hinton's approach, we adapted ConvNeXt by substituting the Gaussian error linear unit with a rectified linear unit and layer normalization with batch normalization. We introduced depth concatenation blocks to fuse information effectively and incorporated a patchify layer. Integrating ConcatNeXt with nested patch-based deep feature engineering, featuring downstream iterative neighborhood component analysis and support vector machine-based functions, establishes a comprehensive approach. ConcatNeXt achieved notable validation and test accuracies of 97.43% and 97.77%, respectively. The ConcatNeXt-based feature engineering model further elevated accuracy to 98.73%. Gradient-weighted class activation maps were employed to provide interpretability, offering valuable insights into model decision-making. Our proposed ConcatNeXt and nested patch-based deep feature engineering models excel in blood cell image classification, showcasing remarkable classification performances. These innovations mark significant strides in computer vision-based blood cell analysis.

Background RNA viruses, including severe acute respiratory syndrome coronavirus-2 (SARS-CoV-2), are important human pathogens. Sequencing of the proteins produced by RNA viruses is essential for understanding disease pathogenesis and may have diagnostic and therapeutic implications. We aimed to develop an accurate and computationally efficient handcrafted feature engineering model for classifying the protein sequences of six pathogenic RNA viruses: SARS-CoV-2, influenza A, influenza B, influenza C, human respirovirus 3, and human immunodeficiency virus (HIV)-1. The first five cause primary respiratory infections; the last has some functional similarity with SARS-CoV-2, justifying the need for diagnostic differentiation. Materials and method We downloaded 14,787 protein sequences belonging to the six categories in FASTA format from the open-source National Center for Biotechnology Information database and transformed the sequences into numeric arrays. First, the signal was divided into overlapping blocks representing three amino acids. Tiny textural motif pattern, a new histogram-based feature extractor, was then applied to extract textural features using simple signum, lower, and upper ternary functions. 512 features were extracted for each protein sequence and fed to an iterative neighborhood component analysis function to select a study dataset-specific optimal number (34) of the most discriminative features for downstream classification using a shallow k-nearest neighbor classifier with 10-fold cross-validation. Novelties: An efficient linear time complexity is introduced for data classification, providing a robust classification approach, especially for complex datasets. Notably, this approach extends beyond the traditional binary classification focus, successfully distinguishing up to six distinct classes. Furthermore, a novel handcrafted feature extraction method is developed, significantly enhancing data analysis and yielding more precise results. Results The model attained 99.71% overall 6-class classification accuracy in a data subset and 99.85% for binary classification of SARS-CoV-2 vs. HIV-1, outperforming a similar published model. Conclusions Our simple model accurately classified the protein sequences of six pathogenic RNA viruses and can potentially be implemented in diagnostic applications to improve RNA virus disease screening.

Abstract Asthma is a common disease. The clinical diagnosis is usually confirmed on a pulmonary function test, which is not always readily accessible. We aimed to develop a computationally lightweight handcrafted machine learning model for asthma detection based on cough sounds recorded using mobile phones. Toward this aim, we proposed a novel feature extractor based on a one-dimensional version of the published attractive-and-repulsive center-symmetric local binary pattern (1D-ARCSLBP), which we tested on a new cough sound dataset. We prospectively recorded cough sounds from 511 asthmatics and 815 non-asthmatic subjects (comprising mostly healthy volunteers), which yielded 1875 one-second cough sound segments for analysis. Our model comprised four steps: (i) preprocessing, in which speech signals and stop times (silent zones between coughs) were removed, leaving behind analyzable cough sound segments; (ii) feature extraction, in which tunable q-factor wavelet transformation was used to perform multilevel signal decomposition into wavelet subbands, allowing 1D-ARCSLBP to extract local low- and high-level features; (iii) feature selection, in which neighborhood component analysis was used to select the most discriminative features; and (iv) classification, in which a standard shallow cubic support vector machine was deployed to calculate binary classification results (asthma versus non-asthma) using tenfold and leave-one-subject-out cross-validations. Our model attained 98.24% and 96.91% accuracy rates with tenfold and leave-one-subject-out cross-validation strategies, respectively, and obtained a low-time complexity. The excellent results confirmed the feature extraction capability of 1D-ARCSLBP and the feasibility of the model being developed into a real-world application for asthma screening.

Abstract In the field of nanoscience, the scanning electron microscope (SEM) is widely employed to visualize the surface topography and composition of materials. In this study, we present a novel SEM image classification model called NFSDense201, which incorporates several key components. Firstly, we propose a unique nested patch division approach that divides each input image into four patches of varying dimensions. Secondly, we utilize DenseNet201, a deep neural network pretrained on ImageNet1k, to extract 2920 deep features from the last fully connected and global average pooling layers. Thirdly, we introduce an iterative neighborhood component analysis function to select the most discriminative features from the merged feature vector, which is formed by concatenating the four feature vectors extracted per input image. This process results in a final feature vector of optimal length 698. Lastly, we employ a standard shallow support vector machine classifier to perform the actual classification. To evaluate the performance of NFSDense201, we conducted experiments using a large public SEM image dataset. The dataset consists of 972, 162, 326, 4590, 3820, 3925, 4755, 181, 917, and 1624.jpeg images belonging to the following microstructural categories: “biological,” “fibers,” “film-coated surfaces,” “MEMS devices and electrodes,” “nanowires,” “particles,” “pattern surfaces,” “porous sponge,” “powder,” and “tips,” respectively. For both four-class and ten-class classification tasks, we evaluated NFSDense201 using subsets of the dataset containing 5080 and 21,272 images, respectively. The results demonstrate the superior performance of NFSDense201, achieving a four-class classification accuracy rate of 99.53% and a ten-class classification accuracy rate of 97.09%. These accuracy rates compare favorably against previously published SEM image classification models. Additionally, we report the performance of NFSDense201 for each class in the dataset.

Abstract Automated urine sediment analysis has become an essential part of diagnosing, monitoring, and treating various diseases that affect the urinary tract and kidneys. However, manual analysis of urine sediment is time-consuming and prone to human bias, and hence there is a need for an automated urine sediment analysis systems using machine learning algorithms. In this work, we propose Swin-LBP, a handcrafted urine sediment classification model using the Swin transformer architecture and local binary pattern (LBP) technique to achieve high classification performance. The Swin-LBP model comprises five phases: preprocessing of input images using shifted windows-based patch division, six-layered LBP-based feature extraction, neighborhood component analysis-based feature selection, support vector machine-based calculation of six predicted vectors, and mode function-based majority voting of the six predicted vectors to generate four additional voted vectors. Our newly reconstructed urine sediment image dataset, consisting of 7 distinct classes, was utilized for training and testing our model. Our proposed model has several advantages over existing automated urinalysis systems. Firstly, we used a feature engineering model that enables high classification performance with linear complexity. This means that it can provide accurate results quickly and efficiently, making it an attractive alternative to time-consuming and biased manual urine sediment analysis. Additionally, our model outperformed existing deep learning models developed on the same source urine sediment image dataset, indicating its superiority in urine sediment classification. Our model achieved 92.60% accuracy for 7-class urine sediment classification, with an average precision of 92.05%. These results demonstrate that the proposed Swin-LBP model can provide a reliable and efficient solution for the diagnosis, surveillance, and therapeutic monitoring of various diseases affecting the kidneys and urinary tract. The proposed model's accuracy, speed, and efficiency make it an attractive option for clinical laboratories and healthcare facilities. In conclusion, the Swin-LBP model has the potential to revolutionize urine sediment analysis and improve patient outcomes in the diagnosis and treatment of urinary tract and kidney diseases.

Cardiac magnetic resonance (CMR) imaging is frequently recommended for patients at intermediate risk of cardiovascular disease to triage them for medication or invasive aggressive treatment. Mitral annulus (MA) motion and velocities represent the cardiac contraction and relaxation, and hold potential to improve the detection of subtle cardiac dysfunction. However, conventional interpretation of CMR images requires expert manipulation and is often operator-dependent. In this paper, we propose an end-to-end MA Point Tracking Network (MANet) to automatically detect and track MA motion during cardiac cycle. The MANet model consists of MA point detection module and motion tracking module. In MA point detection, we design the convolutional-based feature extraction and elastic regression to detect MA points frame by frame of each CMR video. Then, in MA tracking, we adopt the Deep SORT model to capture spatio-temporal continuity between frames and fine-tune the coordinate position of MA points. 171 CMR videos with 4275 frames are used in comparison experiments, and the results demonstrate that our MANet model achieves promising performance in reference to clinical ground truth (r=0.71, P<0.001). This work provides an important preamble for cardiac motion tracking and cardiac function evaluation.

Abstract Coronary artery disease (CAD) is the leading cause of morbidity and death worldwide. Invasive coronary angiography is the most accurate technique for diagnosing CAD, but is invasive and costly. Hence, analytical methods such as machine learning and data mining techniques are becoming increasingly more popular. Although physicians need to know which arteries are stenotic, most of the researchers focus only on CAD detection and few studies have investigated stenosis of the right coronary artery (RCA), left circumflex (LCX) artery and left anterior descending (LAD) artery separately. Meanwhile, most of the datasets in this field are noisy (data uncertainty). However, to the best of our knowledge, there is no study conducted to address this important problem. This study uses the extension of the Z‐Alizadeh Sani dataset, containing 303 records with 54 features. A new feature selection algorithm is proposed in this work. Meanwhile, by discretization of data, we also handle the uncertainty in CAD prediction. To the best of our knowledge, this is the first study attempted to handle uncertainty in CAD prediction. Finally, the genetic algorithm (GA) is used to determine the hyper‐parameters of the support vector machine (SVM) kernels. We have achieved high accuracy for the stenosis diagnosis of each main coronary artery. The results of this study can aid the clinicians to validate their manual stenosis diagnosis of RCA, LCX and LAD coronary arteries.

The quantitative analysis of the left ventricle (LV) contractile function is one of the key steps in the assessment of cardiovascular disease. Such analysis greatly depends on the accurate delineation of LV boundary from cardiac sequences. However, segmentation of the LV still remains a challenging problem due to its subtle boundary, occlusion, and image inhomogeneity. To overcome such difficulties, the authors propose a novel segmentation method by incorporating a dynamic shape constraint into the weighting function of the random walks segmentation algorithm. This approach involves iterative updates on the intermediate result to achieve the desired solution. The inclusion of a shape constraint restricts the solution space of the segmentation result to handle misleading information that may come from noise, weak boundaries and clutter, leading to increased robustness of the algorithm. The authors describe the details of the proposed method and demonstrate its effectiveness in segmenting the LV from real cardiac magnetic resonance (CMR) image sets. The experimental results demonstrate that the proposed method obtains better segmentation performance than the standard method.

Evaluation of right ventricular (RV) structure and function is of importance in the management of most cardiac disorders. But the segmentation of RV has always been considered challenging due to low contrast of the myocardium with surrounding and high shape variability of the RV. In this paper, we present a 2D + T active contour model for segmentation and tracking of RV endocardium on cardiac magnetic resonance (MR) images. To take into account the temporal information between adjacent frames, we propose to integrate the time-dependent constraints into the energy functional of the classical gradient vector flow (GVF). As a result, the prior motion knowledge of RV is introduced in the deformation process through the time-dependent constraints in the proposed GVF-T model. A weighting parameter is introduced to adjust the weight of the temporal information against the image data itself. The additional external edge forces retrieved from the temporal constraints may be useful for the RV segmentation, such that lead to a better segmentation performance. The effectiveness of the proposed approach is supported by experimental results on synthetic and cardiac MR images.