Hybrid Deep Learning Model for Endoscopic Lesion Detection and Classification Using Endoscopy Videos

In medical imaging, the detection and classification of stomach diseases are challenging due to the resemblance of different symptoms, image contrast, and complex background. Computer-aided diagnosis (CAD) plays a vital role in the medical imaging field, allowing accurate results to be obtained in minimal time. This article proposes a new hybrid method to detect and classify stomach diseases using endoscopy videos. The proposed methodology comprises seven significant steps: data acquisition, preprocessing of data, transfer learning of deep models, feature extraction, feature selection, hybridization, and classification. We selected two different CNN models (VGG19 and Alexnet) to extract features. We applied transfer learning techniques before using them as feature extractors. We used a genetic algorithm (GA) in feature selection, due to its adaptive nature. We fused selected features of both models using a serial-based approach. Finally, the best features were provided to multiple machine learning classifiers for detection and classification. The proposed approach was evaluated on a personally collected dataset of five classes, including gastritis, ulcer, esophagitis, bleeding, and healthy. We observed that the proposed technique performed superbly on Cubic SVM with 99.8% accuracy. For the authenticity of the proposed technique, we considered these statistical measures: classification accuracy, recall, precision, False Negative Rate (FNR), Area Under the Curve (AUC), and time. In addition, we provided a fair state-of-the-art comparison of our proposed technique with existing techniques that proves its worthiness.

Download Full-text

Byte2vec: Malware Representation and Feature Selection for Android

The Computer Journal ◽

10.1093/comjnl/bxz121 ◽

2019 ◽

Vol 63 (8) ◽

pp. 1125-1138

Author(s):

Mahmood Yousefi-Azar ◽

Len Hamey ◽

Vijay Varadharajan ◽

Shiping Chen

Keyword(s):

Feature Selection ◽

Feature Vector ◽

False Negative ◽

Malware Detection ◽

False Negative Rate ◽

Selection Algorithm ◽

Feature Selection Algorithm ◽

Context Vector ◽

Wide Range ◽

Selection For

Abstract Malware detection based on static features and without code disassembling is a challenging path of research. Obfuscation makes the static analysis of malware even more challenging. This paper extends static malware detection beyond byte level $n$-grams and detecting important strings. We propose a model (Byte2vec) with the capabilities of both binary file feature representation and feature selection for malware detection. Byte2vec embeds the semantic similarity of byte level codes into a feature vector (byte vector) and also into a context vector. The learned feature vectors of Byte2vec, using skip-gram with negative-sampling topology, are combined with byte-level term-frequency (tf) for malware detection. We also show that the distance between a feature vector and its corresponding context vector provides a useful measure to rank features. The top ranked features are successfully used for malware detection. We show that this feature selection algorithm is an unsupervised version of mutual information (MI). We test the proposed scheme on four freely available Android malware datasets including one obfuscated malware dataset. The model is trained only on clean APKs. The results show that the model outperforms MI in a low-dimensional feature space and is competitive with MI and other state-of-the-art models in higher dimensions. In particular, our tests show very promising results on a wide range of obfuscated malware with a false negative rate of only 0.3% and a false positive rate of 2.0%. The detection results on obfuscated malware show the advantage of the unsupervised feature selection algorithm compared with the MI-based method.

Download Full-text

Tuning IMS station processing parameters and detection thresholds to increase detection precision and decrease detection miss rate

10.5194/egusphere-egu2020-8949 ◽

2020 ◽

Author(s):

Christos Saragiotis ◽

Ivan Kitov

Keyword(s):

False Discovery Rate ◽

Binary Classification ◽

False Negative ◽

Detection Threshold ◽

False Negative Rate ◽

Processing Parameters ◽

Detection Algorithm ◽

Detection Scheme ◽

Statistical Measures ◽

False Discovery

<p>Two principal performance measures of the International Monitoring System (IMS) stations detection capability are the rate of automatic detections associated with events in the Reviewed Event Bulletin (REB) and the rate of detections manually added to the REB. These two metrics roughly correspond to the precision (which is the complement of the false-discovery rate) and miss rate or false-negative rate statistical measures of a binary classification test, respectively. The false-discovery and miss rates are clearly significantly influenced by the number of phases detected by the detection algorithm, which in turn depends on prespecified slowness-, frequency- and azimuth- dependent threshold values used in the short-term average over long-term average ratio detection scheme of the IMS stations. In particular, the lower the threshold, the more the detections and therefore the lower the miss rate but the higher the false discovery rate; the higher the threshold, the less the detections and therefore the higher the miss rate but also the lower the false discovery rate. In that sense decreasing both the false-discovery rate and the miss rate are conflicting goals that need to be balanced. On one hand, it is essential that the miss rate is as low as possible since no nuclear explosion should go unnoticed by the IMS. On the other hand, a high false-discovery rate compromises the quality of the automatically generated event lists and adds heavy and unnecessary workload to the seismic analysts during the interactive processing stage.</p><p>A previous study concluded that a way to decrease both the miss and false-discovery rates as well as the analyst workload is to increase the retiming interval, i.e., the maximum allowable time that an analyst is allowed to move an arrival pick without having to declare a new arrival. Indeed, when a detection needs to be moved by an interval larger than the retiming interval, not only is this a much more time-consuming task for the analyst than just retiming it, but it also affects negatively both the associated rate (the automatic detection is deleted and therefore not associated to an event) and the added rate (a new arrival has to be added to arrival list). The International Data Centre has increased the retiming interval from 4 s to 10 s since October 2018. We show how this change affected the associated-detections and added-detections rates and how the values of these metrics can be further improved by tuning the detection threshold levels.</p>

Download Full-text

Intelligent Personalized Abnormality Detection for Remote Health Monitoring

International Journal of Intelligent Information Technologies ◽

10.4018/ijiit.2020040105 ◽

2020 ◽

Vol 16 (2) ◽

pp. 87-109 ◽

Cited By ~ 1

Author(s):

Poorani Marimuthu ◽

Varalakshmi Perumal ◽

Vaidehi Vijayakumar

Keyword(s):

Machine Learning ◽

Prediction Accuracy ◽

Prediction Models ◽

False Negative ◽

False Negative Rate ◽

Area Under The Curve ◽

Ground Truth ◽

Machine Learning Algorithms ◽

Abnormality Detection ◽

Remote Healthcare

Machine learning algorithms are extensively used in healthcare analytics to learn normal and abnormal patterns automatically. The detection and prediction accuracy of any machine learning model depends on many factors like ground truth instances, attribute relationships, model design, the size of the dataset, the percentage of uncertainty, the training and testing environment, etc. Prediction models in healthcare should generate a minimal false positive and false negative rate. To accomplish high classification or prediction accuracy, the screening of health status needs to be personalized rather than following general clinical practice guidelines (CPG) which fits for an average population. Hence, a personalized screening model (IPAD – Intelligent Personalized Abnormality Detection) for remote healthcare is proposed that tailored to specific individual. The severity level of the abnormal status has been derived using personalized health values and the IPAD model obtains an area under the curve (AUC) of 0.907.

Download Full-text

How to Predict SARS-COV2 Infection in Patient Admitted in Emergency Department: A Prospective Cohort Study.

10.21203/rs.3.rs-122151/v1 ◽

2020 ◽

Author(s):

Frederic BALEN ◽

Sandrine CHAPENTIER ◽

Paul-Henri AUBOIROUX ◽

Elise NOEL-SAVINA ◽

Nicolas SANS ◽

...

Keyword(s):

Respiratory Symptoms ◽

False Negative ◽

False Negative Rate ◽

Area Under The Curve ◽

Unknown Origin ◽

University Hospital ◽

Predictive Tool ◽

Final Model ◽

Negative Rate ◽

Temporal Validation

Abstract Background. In order to rapidly identifying patients with a low probability of being infected by COVID19 to quickly refer them to specialized departments, the objective of our study was to develop a clinical predictive model of infection by COVID19 in patients attending the ED for respiratory symptoms or unexplained fever.Methods. We included all patients over 15 years old, admitted in one of the 2 emergency departments of Toulouse University Hospital between March 13 and March 31 for respiratory symptoms (dyspnea, cough), or fever (or sensation of fever) of unknown origin, and potentially requiring hospitalization. COVID19 infection was assessed by CT-SCAN and RT-PCR. All the candidate predictors were variables collected during the first clinical examination. Internal validation of the final model was performed using the bootstrap procedure. We performed a temporal validation in the same way on patients included between April 1 and April 13.Results. 772 patients were included. The prevalence of COVID19 was 25.5%. There were 19 predictors in the final model. The corrected-by-optimism area under the curve was 0.86 (95%CI = [0.83;0.89]). For a threshold at 10%, the sensitivity was 92%., the specificity was 56%, and the false negative rate was 5%. In secondary data, including 387 patients, the prevalence of COVID19 was 15%. The area under the curve was 0.73 (95%CI = [0.63;0.83]). For the same threshold, the sensitivity was 78%, the specificity was 48%, and the false negative rate was 7%Conclusion. We have developed a predictive tool of COVID19 infection for patients attending the ED. It could safely reduce admission in COVID19 dedicated unit in ED and prevent its overcrowding.Trial registration number: NCT: RC31/20/0149

Download Full-text

The Effect of a Diverse Dataset for Transfer Learning in Thermal Person Detection

Sensors ◽

10.3390/s20071982 ◽

2020 ◽

Vol 20 (7) ◽

pp. 1982 ◽

Cited By ~ 1

Author(s):

Noor Ul Huda ◽

Bolette D. Hansen ◽

Rikke Gade ◽

Thomas B. Moeslund

Keyword(s):

Transfer Learning ◽

Privacy Preservation ◽

False Negative ◽

False Negative Rate ◽

Model Adaptation ◽

True Negative ◽

True Negative Rate ◽

Negative Rate ◽

Thermal Cameras ◽

The Impact

Thermal cameras are popular in detection for their precision in surveillance in the dark and for privacy preservation. In the era of data driven problem solving approaches, manually finding and annotating a large amount of data is inefficient in terms of cost and effort. With the introduction of transfer learning, rather than having large datasets, a dataset covering all characteristics and aspects of the target place is more important. In this work, we studied a large thermal dataset recorded for 20 weeks and identified nine phenomena in it. Moreover, we investigated the impact of each phenomenon for model adaptation in transfer learning. Each phenomenon was investigated separately and in combination. the performance was analyzed by computing the F1 score, precision, recall, true negative rate, and false negative rate. Furthermore, to underline our investigation, the trained model with our dataset was further tested on publicly available datasets, and encouraging results were obtained. Finally, our dataset was also made publicly available.

Download Full-text

Diagnostic accuracy of fetal scalp lactate for intrapartum acidosis compared with scalp pH

Journal of Perinatal Medicine ◽

10.1515/jpm-2016-0044 ◽

2017 ◽

Vol 45 (3) ◽

Author(s):

Jara Pascual Mancho ◽

Sabina Marti Gamboa ◽

Olga Redrado Gimenez ◽

Raquel Crespo Esteras ◽

Belen Rodriguez Solanilla ◽

...

Keyword(s):

Metabolic Acidosis ◽

Diagnostic Accuracy ◽

False Negative ◽

False Negative Rate ◽

Area Under The Curve ◽

Predictive Ability ◽

Blood Sampling ◽

Obstetric Management ◽

Fetal Scalp ◽

Fetal Scalp Blood Sampling

AbstractObjective:To determine the diagnostic accuracy of fetal scalp lactate sampling (FSLS) and to establish an optimal cut-off value for intrapartum acidosis compared with fetal scalp pH.Methods:A 20-month retrospective cohort study was conducted of all neonates delivered in our institution for whom fetal scalp blood sampling (FSBS) was performed, matching their intrapartum gasometry to their cord gasometry at delivery (n=243). The time taken from the performance of scalp blood sampling to arterial umbilical cord gas acquisition was 45 min at most. Five arterial cord gasometry patterns were set for assessing the predictive ability of both techniques. Subsequent obstetric management for a pathological value was analysed considering the use of both techniques.Results:The optimal cut-off value for FSLS was 4.8 mmol/L: this value has 100% sensitivity and 63% specificity for umbilical arterial cord gas pH≤7.0 and base deficit (BD)≥12 detection, and 100% sensitivity and 64% specificity for umbilical arterial cord gas pH≤7.10 and BD≥12 detection, with a false negative rate of <1.3%, improving fetal scalp pH performance. FSLS showed the best area under the curve (AUC) of 0.86 and 0.84 for both arterial cord gasometry patterns, respectively. Expedite birth following lactate criteria would have been the same as following pH criteria (92 obstetric interventions) with no cases of missed metabolic acidosis. In the cohort, 19.8% of cases were discordant, but no cases of metabolic acidosis were in this group.Conclusions:FSLS improves the detection of metabolic acidosis via fetal scalp pH with an optimal cut-off value of 4.8 mmol/L. FSLS can be used without increasing obstetrical interventions or missing metabolic acidosis.

Download Full-text

Classification of Biodegradable Substances Using Balanced Random Trees and Boosted C5.0 Decision Trees

International Journal of Environmental Research and Public Health ◽

10.3390/ijerph17249322 ◽

2020 ◽

Vol 17 (24) ◽

pp. 9322

Author(s):

Alaa M. Elsayad ◽

Ahmed M. Nassef ◽

Mujahed Al-Dhaifallah ◽

Khaled A. Elsayad

Keyword(s):

Decision Trees ◽

Roc Curve ◽

Molecular Descriptor ◽

False Positive Rate ◽

False Negative ◽

False Negative Rate ◽

Random Trees ◽

Support Vector ◽

Statistical Measures ◽

Qsar Models

Substances that do not degrade over time have proven to be harmful to the environment and are dangerous to living organisms. Being able to predict the biodegradability of substances without costly experiments is useful. Recently, the quantitative structure–activity relationship (QSAR) models have proposed effective solutions to this problem. However, the molecular descriptor datasets usually suffer from the problems of unbalanced class distribution, which adversely affects the efficiency and generalization of the derived models. Accordingly, this study aims at validating the performances of balanced random trees (RTs) and boosted C5.0 decision trees (DTs) to construct QSAR models to classify the ready biodegradation of substances and their abilities to deal with unbalanced data. The balanced RTs model algorithm builds individual trees using balanced bootstrap samples, while the boosted C5.0 DT is modeled using cost-sensitive learning. We employed the two-dimensional molecular descriptor dataset, which is publicly available through the University of California, Irvine (UCI) machine learning repository. The molecular descriptors were ranked according to their contributions to the balanced RTs classification process. The performance of the proposed models was compared with previously reported results. Based on the statistical measures, the experimental results showed that the proposed models outperform the classification results of the support vector machine (SVM), K-nearest neighbors (KNN), and discrimination analysis (DA). Classification measures were analyzed in terms of accuracy, sensitivity, specificity, precision, false positive rate, false negative rate, F1 score, receiver operating characteristic (ROC) curve, and area under the ROC curve (AUROC).

Download Full-text

Utility of SNOMED CT in automated expansion of clinical terms in discharge summaries: Testing issues of coverage

Health Information Management Journal ◽

10.1177/1833358320934528 ◽

2020 ◽

pp. 183335832093452

Author(s):

Aleksandar Zivaljevic ◽

Koray Atalag ◽

James Warren

Keyword(s):

False Negative ◽

False Negative Rate ◽

True Positive Rate ◽

Knowledge Extraction ◽

Snomed Ct ◽

Automated Method ◽

Statistical Measures ◽

Rest Api ◽

The Individual ◽

Discharge Summaries

Objective: This study tests coverage of SNOMED CT as an expansion source in the process of automated expansion of clinical terms found in discharge summaries. Term expansion is commonly used as a technique in knowledge extraction, query formulation and semantic modelling among other applications. However, characteristics of the sources might affect credibility of outputs, and coverage is one of them. Method: We developed an automated method for testing coverage of more than one source at a time. We used several methods to clean our corpus of discharge summaries before we extracted text fragments as candidates for clinical concepts. We then used Unified Medical Language System (UMLS) sources and UMLS REST API to filter concepts from the pool of text fragments. Statistical measures like true positive rate and false negative rate were used to decide on the coverage of the source. We also tested the coverage of the individual SNOMED CT hierarchies using the same methods. Results: Findings suggest that a combination of four terminologies tested (SNOMED CT, NCI, LNC and MSH) achieves over 90% of coverage for term expansion. We also found that the SNOMED CT hierarchies that hold clinically relevant concepts provided 60% of coverage. Conclusion: We believe that our findings and the method we developed will be of use to both scientists and practitioners working in the domain of knowledge extraction.

Download Full-text

Clinically Meaningful Change

Methodology ◽

10.1027/1614-2241/a000168 ◽

2019 ◽

Vol 15 (3) ◽

pp. 97-105

Author(s):

Rodrigo Ferrer ◽

Antonio Pardo

Keyword(s):

Effect Size ◽

False Negative ◽

False Negative Rate ◽

Point Of View ◽

Skewed Distribution ◽

Effect Sizes ◽

False Negatives ◽

Large Size ◽

Before And After ◽

Post Test

Abstract. In a recent paper, Ferrer and Pardo (2014) tested several distribution-based methods designed to assess when test scores obtained before and after an intervention reflect a statistically reliable change. However, we still do not know how these methods perform from the point of view of false negatives. For this purpose, we have simulated change scenarios (different effect sizes in a pre-post-test design) with distributions of different shapes and with different sample sizes. For each simulated scenario, we generated 1,000 samples. In each sample, we recorded the false-negative rate of the five distribution-based methods with the best performance from the point of view of the false positives. Our results have revealed unacceptable rates of false negatives even with effects of very large size, starting from 31.8% in an optimistic scenario (effect size of 2.0 and a normal distribution) to 99.9% in the worst scenario (effect size of 0.2 and a highly skewed distribution). Therefore, our results suggest that the widely used distribution-based methods must be applied with caution in a clinical context, because they need huge effect sizes to detect a true change. However, we made some considerations regarding the effect size and the cut-off points commonly used which allow us to be more precise in our estimates.

Download Full-text

Diagnostic Value of a Carpal Tunnel Corticosteroid Injection in Patients with Negative Electrodiagnostic Studies

Journal of Hand and Microsurgery ◽

10.1055/s-0040-1717830 ◽

2020 ◽

Author(s):

Brian M. Katt ◽

Casey Imbergamo ◽

Fortunato Padua ◽

Joseph Leider ◽

Daniel Fletcher ◽

...

Keyword(s):

Carpal Tunnel ◽

Corticosteroid Injection ◽

Carpal Tunnel Release ◽

False Negative ◽

False Negative Rate ◽

Signs And Symptoms ◽

Diagnostic Value ◽

Symptom Improvement ◽

Study Cohort ◽

Chi Square Analysis

Abstract Introduction There is a known false negative rate when using electrodiagnostic studies (EDS) to diagnose carpal tunnel syndrome (CTS). This can pose a management dilemma for patients with signs and symptoms that correlate with CTS but normal EDS. While corticosteroid injection into the carpal tunnel has been used in this setting for diagnostic purposes, there is little data in the literature supporting this practice. The purpose of this study is to evaluate the prognostic value of a carpal tunnel corticosteroid injection in patients with a normal electrodiagnostic study but exhibiting signs and symptoms suggestive of carpal tunnel, who proceed with a carpal tunnel release. Materials and Methods The group included 34 patients presenting to an academic orthopedic practice over the years 2010 to 2019 who had negative EDS, a carpal tunnel corticosteroid injection, and a carpal tunnel release. One patient (2.9%), where the response to the corticosteroid injection was not documented, was excluded from the study, yielding a study cohort of 33 patients. Three patients had bilateral disease, yielding 36 hands for evaluation. Statistical analysis was performed using Chi-square analysis for nonparametric data. Results Thirty-two hands (88.9%) demonstrated complete or partial relief of neuropathic symptoms after the corticosteroid injection, while four (11.1%) did not experience any improvement. Thirty-one hands (86.1%) had symptom improvement following surgery, compared with five (13.9%) which did not. Of the 32 hands that demonstrated relief following the injection, 29 hands (90.6%) improved after surgery. Of the four hands that did not demonstrate relief after the injection, two (50%) improved after surgery. This difference was statistically significant (p = 0.03). Conclusion Patients diagnosed with a high index of suspicion for CTS do well with operative intervention despite a normal electrodiagnostic test if they have had a positive response to a preoperative injection. The injection can provide reassurance to both the patient and surgeon before proceeding to surgery. Although patients with a normal electrodiagnostic test and no response to cortisone can still do well with surgical intervention, the surgeon should carefully review both the history and physical examination as surgical success may decrease when both diagnostic tests are negative. Performing a corticosteroid injection is an additional diagnostic tool to consider in the management of patients with CTS and normal electrodiagnostic testing.

Download Full-text