BioExpDNN: Bioinformatic Explainable Deep Neural Network

Classification of Bisyllabic Lexical Stress Patterns Using Deep Neural Networks

The present work strives to investigate the effect of using dimensionality reduction techniques (DRTs) on breast cancer (BC) classification problem. Primarily, we focused on the following five (DRTs): Auto-Encoders (AE), T-Distributed Stochastic Neighbor Embedding (T- SNE), Recursive Feature Elimination (RFE), Isometric Feature Mapping (Isomap), and Principle Component analysis (PCA). These methods are combined with two famous classifiers that are Support Vector Machine (SVM) and Multilayer perceptron (MLP). They are used for BC classification. Breast Cancer Wisconsin Diagnostic (WDBC) data set was used to validate the experiments of this work. The former was provided by the University of California, Irvine (UCI) machine learning repository. The results demonstrated that combining MLP with the chosen (DRTs) methods increased the classification accuracy for almost all built models by at least 0.7%. In addition, they revealed a decrease in the classification accuracy using SVM as a classifier for almost all built models.KeywordsMachine learningDeep learningDimensionality reductionNeural networksBreast cancer predictionMedical dataset

Heartbeat classification is a crucial step for arrhythmia diagnosis during electrocardiographic (ECG) analysis. The new scenario of wireless body sensor network- (WBSN-) enabled ECG monitoring puts forward a higher-level demand for this traditional ECG analysis task. Previously reported methods mainly addressed this requirement with the applications of a shallow structured classifier and expert-designed features. In this study, modified frequency slice wavelet transform (MFSWT) was firstly employed to produce the time-frequency image for heartbeat signal. Then the deep learning (DL) method was performed for the heartbeat classification. Here, we proposed a novel model incorporating automatic feature abstraction and a deep neural network (DNN) classifier. Features were automatically abstracted by the stacked denoising auto-encoder (SDA) from the transferred time-frequency image. DNN classifier was constructed by an encoder layer of SDA and a softmax layer. In addition, a deterministic patient-specific heartbeat classifier was achieved by fine-tuning on heartbeat samples, which included a small subset of individual samples. The performance of the proposed model was evaluated on the MIT-BIH arrhythmia database. Results showed that an overall accuracy of 97.5% was achieved using the proposed model, confirming that the proposed DNN model is a powerful tool for heartbeat pattern recognition.

Aim To estimate the incremental phase-specific and lifetime economic burden among newly diagnosed cervical and endometrial cancer patients versus non-cancer controls. Methods Cervical and endometrial cancer patients newly diagnosed between January 2015 and June 2018 were identified in the Optum Clinformatics DataMart database. The index date was the date of the first diagnosis for cancer cases and the first claim date after 12 months of continuous enrollment for non-cancer controls. Patients were followed until death/loss of enrollment/end of data availability. Per patient per month (PPPM) costs attributable to cancer were calculated for 4 phases: pre-diagnosis (3 months prior to diagnosis), initial (6 months post-diagnosis), terminal (6 months pre-death), and continuation (remaining time between initial and terminal phases). Survival data were obtained to determine the monthly proportion of patients in each phase. Total survival adjusted monthly costs were obtained by multiplying the proportion of patients in each phase by the total cost incurred during that month. Phase-specific and lifetime incremental costs of cervical and endometrial cancer were obtained using generalized linear models. Results The analytic cohort included 1,002 cervical cancer patients and 4,005 matched non-cancer controls and 5,003 endometrial cancer patients matched with 19,999 non-cancer controls. Mean adjusted incremental PPPM lifetime costs (95% CI) for cervical cancer and endometrial cancer cases were $5,910 ($5,373-$6,446) and $3,475 ($3,259-$3,691), respectively. Incremental total PPPM phase-specific costs attributable to cervical and endometrial cancer were pre-diagnosis (cervical: $1,057; endometrial: $3,315), initial ($12,084; $8,618), continuation ($2,732; $1,147), and terminal ($2,702; $5,442). Incremental costs were significantly higher for cancer patients versus non-cancer controls across patient lifetime and all phases of care (except terminal phase costs for cervical cancer). Outpatient costs were the major driver of costs across all post-diagnosis phases. Conclusion This study highlights the cost burden associated with cervical/endometrial cancer and cost variation by phases of care.

Wearable inertial measurement unit (IMU) sensors are powerful enablers for acquisition of motion data. Specifically, in human activity recognition (HAR), IMU sensor data collected from human motion are categorically combined to formulate datasets that can be used for learning human activities. However, successful learning of human activities from motion data involves the design and use of proper feature representations of IMU sensor data and suitable classifiers. Furthermore, the scarcity of labelled data is an impeding factor in the process of understanding the performance capabilities of data-driven learning models. To tackle these challenges, two primary contributions are in this article: first; by using raw IMU sensor data, a spectrogram-based feature extraction approach is proposed. Second, an ensemble of data augmentations in feature space is proposed to take care of the data scarcity problem. Performance tests were conducted on a deep long term short term memory (LSTM) neural network architecture to explore the influence of feature representations and the augmentations on activity recognition accuracy. The proposed feature extraction approach combined with the data augmentation ensemble produces state-of-the-art accuracy results in HAR. A performance evaluation of each augmentation approach is performed to show the influence on classification accuracy. Finally, in addition to using our own dataset, the proposed data augmentation technique is evaluated against the University of California, Irvine (UCI) public online HAR dataset and yields state-of-the-art accuracy results at various learning rates.

BackgroundProbabilistic assessments of clinical care are essential for quality care. Yet, machine learning, which supports this care process has been limited to categorical results. To maximize its usefulness, it is important to find novel approaches that calibrate the ML output with a likelihood scale. Current state-of-the-art calibration methods are generally accurate and applicable to many ML models, but improved granularity and accuracy of such methods would increase the information available for clinical decision making.This novel non-parametric Bayesian approach is demonstrated on a variety of data sets, including simulated classifier outputs, biomedical data sets from the University of California, Irvine (UCI) Machine Learning Repository, and a clinical data set built to determine suicide risk from the language of emergency department patients.ResultsThe method is first demonstrated on support-vector machine (SVM) models, which generally produce well-behaved, well understood scores. The method produces calibrations that are comparable to the state-of-the-art Bayesian Binning in Quantiles (BBQ) method when the SVM models are able to effectively separate cases and controls. However, as the SVM models’ ability to discriminate classes decreases, our approach yields more granular and dynamic calibrated probabilities comparing to the BBQ method. Improvements in granularity and range are even more dramatic when the discrimination between the classes is artificially degraded by replacing the SVM model with an ad hoc k-means classifier.ConclusionsThe method allows both clinicians and patients to have a more nuanced view of the output of an ML model, allowing better decision making. The method is demonstrated on simulated data, various biomedical data sets and a clinical data set, to which diverse ML methods are applied. Trivially extending the method to (non-ML) clinical scores is also discussed.

Introduction: Cervical cancer (CC) is a prevailing malignant tumor in women, mainly caused by human papillomavirus (HPV) infection. This study investigated miR-106a expression in the serum of HPV-positive CC patients and estimated its value in diagnosis and prognosis. Methods: We enrolled 120 CC patients as study subjects, with another 80 healthy women as controls. Clinical baseline data and clinicopathological indexes including age, tumor size, differentiation degree, FIGO stage, lymph node metastasis, and squamous cell carcinoma antigen (SCC-Ag) were recorded. Serum miR-106a expression was measured using reverse transcription-quantitative polymerase chain reaction. Receiver operating characteristic curve was employed to estimate the efficacy of miR-106a in diagnosing CC or HPV-positive CC. Under a 5-year follow-up, patient survival was recorded, and the impact of miR-106a on overall survival rate was analyzed by the Kaplan-Meier method. The logistic regression model was used to analyze whether miR-106a was an independent prognostic factor for HPV infection in CC patients. Results: Serum miR-106a was upregulated in CC patients and the level >1.365 assisted the CC diagnosis. miR-106a expression in HPV-positive CC patients was elevated relative to HPV-negative CC patients, and serum miR-106a level >1.300 distinguishing HPV positive and HPV negative. HPV positivity was linked with tumor differentiation degree, FIGO stage, lymph node metastasis, and SCC-Ag in CC patients, but not with age and tumor size. High expression of miR-106a in HPV-positive CC patients increased the risk of poor prognosis, and miR-106a expression is an independent prognostic factor for HPV infection in CC patients. Conclusion: High expression of miR-106a assists in the diagnosis of HPV-positive CC and predicts poor prognosis.

There are some problems in the current human motion target gesture recognition algorithms, such as classification accuracy, overlap ratio, low recognition accuracy and recall, and long recognition time. A gesture recognition algorithm of human motion based on deep neural network was proposed. First, Kinect interface equipment was used to collect the coordinate information of human skeleton joints, extract the characteristics of motion gesture nodes, and construct the whole structure of key node network by using deep neural network. Second, the local recognition region was introduced to generate high-dimensional feature map, and the sampling kernel function was defined. The minimum space-time domain of node structure map was located by sampling in the space-time domain. Finally, the deep neural network classifier was constructed to integrate and classify the human motion target gesture data features to realize the recognition of human motion target. The results show that the proposed algorithm has high classification accuracy and overlap ratio of human motion target gesture, the recognition accuracy is as high as 93%, the recall rate is as high as 88%, and the recognition time is 17.8 s, which can effectively improve the human motion target attitude recognition effect.

Deep neural networks (DNNs), which learn a hierarchical representation of features, have shown remarkable performance in big data analytics of remote sensing. However, previous research indicates that DNNs are easily spoofed by adversarial examples, which are crafted images with artificial perturbations that fool DNN models toward wrong predictions. To comprehensively evaluate the impact of adversarial examples on the remote sensing image (RSI) scene classification, this study tests eight state-of-the-art classification DNNs on six RSI benchmarks. These data sets include both optical and synthetic-aperture radar (SAR) images of different spectral and spatial resolutions. In the experiment, we create 48 classification scenarios and use four cutting-edge attack algorithms to investigate the influence of the adversarial example on the classification of RSIs. The experimental result shows that the fooling rates of the attacks are all over 98% across the 48 scenarios. We also find that, for the optical data, the seriousness of the adversarial problem has a negative relationship with the richness of the feature information. Besides, adversarial examples generated from SAR images are used easily for fooling the models with an average fooling rate of 76.01%. By analyzing the class distribution of these adversarial examples, we find that the distribution of the misclassifications is not affected by the types of models and attack algorithms-adversarial examples of RSIs of the same class cluster on fixed several classes. The analysis of classes of adversarial examples not only helps us explore the relationships between data set classes but also provides insights for further designing defensive algorithms.

Analysis and classification of sequences is one of the key research areas in bioinformatics. The basic tool for sequence analysis is alignment, but there are also other techniques that can be used. Frequency Chaos Game Representation is a technique that builds an image characteristic of the sequence The paper describes the first experiment in the use of a deep neural network for classification of DNA sequences represented as images by using the Frequency Chaos Game Representation.

ObjectThis study aimed to establish an effective risk nomogram to predict the early distant metastasis (EDM) probability of cervical cancer (CC) patients treated with radical radiotherapy to aid individualized clinical decision‐making.MethodsA total of 489 patients with biopsy‐confirmed CC between December 2018 and January 2021 were enrolled. Logistic regression with the stepwise backward method was used to identify independent risk factors. The nomogram efficacy was evaluated by using the area under the receiver operating characteristic curve (AUC), C‐index by 1000 bootstrap replications, etc. Finally, patients were divided into high‐ and low‐risk groups of EDM based on the cut‐off value of nomogram points.Results36 (7.36%) CC patients had EDM, and 20 (55.6%) EDM had more than one metastatic site involved. Age below 51 (OR = 2.298, p < 0.001), tumor size larger than 4.5 cm (OR = 3.817, p < 0.001) and radiotherapy (OR = 3.319, p < 0.001) were independent risk factors of EDM. For the nomogram model, C‐index was 0.701 (95% CI = 0.604–0.798), and 0.675 (95% CI = 0.578–0.760) after 1000 bootstrap resampling validations. The Hosmer–Lemeshow test demonstrated no overfitting (p = 0.924). According to the Kaplan–Meier curve of risk score, patients with high risk were more prone to get EDM (p < 0.001).ConclusionThis is the first research to focus on EDM in CC patients. We have developed a robust scoring system to predict the risk of EDM in CC patients to screen out appropriate cases for consolidation therapy and more intensive follow‐up.

Cervical cancer (CC) is a common female malignancy, with a global incidence rate second only to breast cancer. To propose a new idea for early treatment and auxiliary diagnosis of CC by exploring the diagnostic and prognostic implications of serum miR-182 in CC. We enrolled 70 CC patients, 35 cervical intraepithelial neoplasia (CIN) patients and 35 healthy controls (HCs), who visited The First Affiliated Hospital of Hainan Medical College Hospital between January 2015 and April 2016. miR-182 expression was quantified by real-time quantitative PCR and compared among the three groups. The correlation of serum miR-182 expression with patients' clinical features was evaluated. The receiver operating characteristic curve (ROC) and the Kaplan-Meier method were used to evaluate the early diagnostic value and prognostic value of serum miR-182. Cox regression analysis was performed to determine serum miR-182 expression and its important role in predicting CC patients' prognosis. Serum miR-182 expression was determined to be 0.345 ± 0.094, 0.369 ± 0.076, and 0.586 ± 0.157 in CC patients, CIN patients, and HCs, respectively (P< 0.001). Serum miR-182 expression had an obvious association with lymph node metastasis and pathological differentiation (P< 0.05). The area under the ROC curve (AUC) of serum miR-182 was 0.709 (95% CI: 0.622-0.795), the critical value was 0.456, the sensitivity was 81.4%, and the specificity was 52.9%. CC patients were grouped as either the low- (miR-182 < 0.3) or high-level group (miR-182 ⩾ 0.03) based on serum miR-182 levels, and a Cox regression model of OS was established. Serum miR-182 expression was identified as a factor independently influencing CC patients' OS (P= 0.028); the death risk of the high-level group was 3.246 times that of the low-level group. Serum miR-182 expression is not only a biomarker for early diagnosis of CC, but also one of the independent factors influencing the survival and prognosis of CC patients.

Background: In this study, we used scanning electron microscopy (SEM) and atomic force microscopy (AFM) to examine the changes in morphology of red blood cells (RBCs) and to investigate the nanoparticles (NPs) found on their surface in cervical cancer (CC) patients undergoing radiation therapy (RT). Methods and Results: We obtained smears of venous blood from 12 CC patients at the start, midway and at the end of external beam RT and then midway and at the end of brachytherapy. It was found that in CC patients, the number of RBCs with abnormal morphology increased and NPs appeared on their surface. During RT, the total number of abnormally shaped RBCs and the number and size of NPs increased. The RBC diameter was 8.38 +/- 0.36 mu m in the control group and 9.41 +/- 0.47 mu m in CC patients. The average diameter of NPs on the RBC surface was 69.91 +/- 12.15 nm and their average height 23.75 +/- 3.70 nm. After RT, the morphology of RBCs was restored, and the formation of NPs decreased. Conclusion: The changes observed could serve as the basis for developing efficacy indicators of cancer radiation therapy.

Accumulated evidence has suggested a relatively high recurrence rate in early-stage cervical cancer (CC) patients with risk factors. This study aimed to assess the efficacy and safety of consolidation chemotherapy following adjuvant therapy (concurrent chemoradiotherapy (CCRT) or radiotherapy (RT) alone) in stage IB-IIA CC patients with risk factors. A total of 237 stage IB-IIA CC patients who received radical surgery between January 2014 and December 2021 were included in the retrospective study. According to the types of adjuvant therapies, the patients were classified into the control group (CCRT or RT alone) and the study group (consolidation chemotherapy following CCRT or RT alone). The propensity score matching (PSM) was used to balance baseline characteristics between the two groups. The primary end points of the study were disease-free survival (DFS) and overall survival (OS). For the entire cohort, no significant difference was observed in the DFS or OS between the study and control group, which was also confirmed in the PSM cohort (n=124). The multivariate analysis identified the high-risk factor type was an independent adverse prognostic factor for the patients. In patients with high risk factors, consolidation chemotherapy following adjuvant therapy was significantly associated with better clinical outcomes and identified as an independent prognostic favorable factor. Moreover, this association remained statistically significant in high-risk patients with ≥2 metastatic lymph nodes. In patients with intermediate risk factors, consolidation chemotherapy following adjuvant therapy was unrelated to DFS or OS. The safe assessment demonstrated consolidation chemotherapy following adjuvant therapy was significantly correlated with higher rates of ≥ grade 3 hematologic toxicities in both the global and subgroup analysis stratified by risk factor type. Consolidation chemotherapy after adjuvant therapy provided survival benefits in stage IB-IIA CC patients with high risk factors, particularly those with ≥2 metastatic lymph nodes. However, related hematologic toxicities should be alerted in patient management. The actual efficacy and safety of consolidation chemotherapy still need to be investigated in more well-designed clinical trials.

Introduction/BackgroundThe purpose of the study was a comparative analysis of the intensity of lipid peroxidation (LPO) and antioxidant system in the blood of cervical cancer (CC) patients receiving treatment with...