Deep learning model for distributed denial of service (DDoS) detection

Distributed denial of service (DDoS) attacks is one of the serious threats in the domain of cybersecurity where it affects the availability of online services by disrupting access to its legitimate users. The consequences of such attacks could be millions of dollars in worth since all of the online services are relying on high availability. The magnitude of DDoS attacks is ever increasing as attackers are smart enough to innovate their attacking strategies to expose vulnerabilities in the intrusion detection models or mitigation mechanisms. The history of DDoS attacks reflects that network and transport layers of the OSI model were the initial target of the attackers, but the recent history from the cybersecurity domain proves that the attacking momentum has shifted toward the application layer of the OSI model which presents a high degree of difficulty distinguishing the attack and benign traffics that make the combat against application-layer DDoS attack a sophisticated task. Striding for high accuracy with high DDoS classification recall is key for any DDoS detection mechanism to keep the reliability and trustworthiness of such a system. In this paper, a deep learning approach for application-layer DDoS detection is proposed by using an autoencoder to perform the feature selection and Deep neural networks to perform the attack classification. A popular benchmark dataset CIC DoS 2017 is selected by extracting the most appealing features from the packet flows. The proposed model has achieved an accuracy of 99.83% with a detection rate of 99.84% while maintaining the false-negative rate of 0.17%, which has the heights accuracy rate among the literature reviewed so far.

Spontaneous mutation rate estimates for the principal malaria vectors Anopheles coluzzii and Anopheles stephensi

Using high-depth whole genome sequencing of F0 mating pairs and multiple individual F1 offspring, we estimated the nuclear mutation rate per generation in the malaria vectors Anopheles coluzzii and Anopheles stephensi by detecting de novo genetic mutations. A purpose-built computer program was employed to filter actual mutations from a deep background of superficially similar artifacts resulting from read misalignment. Performance of filtering parameters was determined using software-simulated mutations, and the resulting estimate of false negative rate was used to correct final mutation rate estimates. Spontaneous mutation rates by base substitution were estimated at 1.00 × 10−9 (95% confidence interval, 2.06 × 10−10—2.91 × 10−9) and 1.36 × 10−9 (95% confidence interval, 4.42 × 10−10—3.18 × 10−9) per site per generation in A. coluzzii and A. stephensi respectively. Although similar studies have been performed on other insect species including dipterans, this is the first study to empirically measure mutation rates in the important genus Anopheles, and thus provides an estimate of µ that will be of utility for comparative evolutionary genomics, as well as for population genetic analysis of malaria vector mosquito species.

Limited-angle TOF-PET for intraoperative surgical applications: proof of concept and first experimental data

The intraoperative gamma probe (IPG) based on single gamma-ray detection remains the current gold standard modality for sentinel lymph node identification and tumor removal in cancer patients. However, IPGs do not meet the <5% false negative rate (FNR) requirement, a key metric suggested by the American Society of Clinical Oncology (ASCO). We aim to reduce FNR by using time of flight (TOF) PET detector technology in limited angle geometry system by using only two detector panels in coincidence. For proof of concept, we used two Hamamatsu TOF PET detector modules (C13500-4075YC-12) featuring 12× 12 arrays of 4.14× 4.14× 20 mm3 LFS crystal pixels with 4.2 mm pitch and coupled one-one to silicon photomultiplier (SiPM) pixels. The measured detector coincidence timing resolution (CTR) was 271 ps FWHM for the whole detector. We 3D printed lesion phantom containing spheres 2–10 mm in diameter, representing lymph nodes, and placed it inside a 10-liter warm background water phantom. Experimental results showed that with subminute data acquisition, 6 mm diameter spheres could be identified in the image when a lesion phantom with a 10:1 activity ratio to background was used. The simulation results were in good agreement with the experimental data by resolving 6 mm diameter spherical lesions with a 60 second acquisition time in a 25 cm deep background water phantom with a 10:1 activity ratio. As expected, the image quality improved as the CTR improved in the simulation and with decreasing background water phantom depth or increasing lesion-to-background activity ratio in the experiment. With the results presented here, we concluded that using a limited angle TOF PET detector system is a major step forward for intraoperative applications in that lesion detectability is beyond what conventional gamma- and NIR-based probes could achieve.

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.

Machine Learning Assisted Cervical Cancer Detection

Cervical malignant growth is the fourth most typical reason for disease demise in women around the globe. Cervical cancer growth is related to human papillomavirus (HPV) contamination. Early screening made cervical cancer a preventable disease that results in minimizing the global burden of cervical cancer. In developing countries, women do not approach sufficient screening programs because of the costly procedures to undergo examination regularly, scarce awareness, and lack of access to the medical center. In this manner, the expectation of the individual patient's risk becomes very high. There are many risk factors relevant to malignant cervical formation. This paper proposes an approach named CervDetect that uses machine learning algorithms to evaluate the risk elements of malignant cervical formation. CervDetect uses Pearson correlation between input variables as well as with the output variable to pre-process the data. CervDetect uses the random forest (RF) feature selection technique to select significant features. Finally, CervDetect uses a hybrid approach by combining RF and shallow neural networks to detect Cervical Cancer. Results show that CervDetect accurately predicts cervical cancer, outperforms the state-of-the-art studies, and achieved an accuracy of 93.6%, mean squared error (MSE) error of 0.07111, false-positive rate (FPR) of 6.4%, and false-negative rate (FNR) of 100%.

Understanding the Practices of Global Censorship through Accurate, End-to-End Measurements

It is challenging to conduct a large scale Internet censorship measurement, as it involves triggering censors through artificial requests and identifying abnormalities from corresponding responses. Due to the lack of ground truth on the expected responses from legitimate services, previous studies typically require a heavy, unscalable manual inspection to identify false positives while still leaving false negatives undetected. In this paper, we propose Disguiser, a novel framework that enables end-to-end measurement to accurately detect the censorship activities and reveal the censor deployment without manual efforts. The core of Disguiser is a control server that replies with a static payload to provide the ground truth of server responses. As such, we send requests from various types of vantage points across the world to our control server, and the censorship activities can be recognized if a vantage point receives a different response. In particular, we design and conduct a cache test to pre-exclude the vantage points that could be interfered by cache proxies along the network path. Then we perform application traceroute towards our control server to explore censors' behaviors and their deployment. With Disguiser, we conduct 58 million measurements from vantage points in 177 countries. We observe 292 thousand censorship activities that block DNS, HTTP, or HTTPS requests inside 122 countries, achieving a 10^-6 false positive rate and zero false negative rate. Furthermore, Disguiser reveals the censor deployment in 13 countries.

An Exploration into the Detection of COVID-19 from Chest X-ray Scans Using the xRGM-NET Convolutional Neural Network

COVID-19 has spread rapidly across the world since late 2019. As of December, 2021, there are over 250 million documented COVID-19 cases and over 5 million deaths worldwide, which have caused businesses, schools, and government operations to shut down. The most common method of detecting COVID-19 is the RT-PCR swab test, which suffers from a high false-negative rate and a very slow turnaround for results, often up to two weeks. Because of this, specialists often manually review X-ray images of the lungs to detect the presence of COVID-19 with up to 97% accuracy. Neural network algorithms greatly accelerate this review process, analyzing hundreds of X-rays in seconds. Using the Cohen COVID-19 X-ray Database and the NIH ChestX-ray8 Database, we trained and constructed the xRGM-NET convolutional neural network (CNN) to detect COVID-19 in X-ray scans of the lungs. To further aid medical professionals in the manual review of X-rays, we implemented the CNN activation mapping technique Score-CAM, which generates a heat map over an X-ray to illustrate which areas in the scan are most influential over the ultimate diagnosis. xRGM-NET achieved an overall classification accuracy of 97% with a sensitivity of 94% and specificity of 97%. Lightweight models like xRGM-NET can serve to improve the efficiency and accuracy of COVID-19 detection in developing countries or rural areas. In this paper, we report on our model and methods that were developed as part of a STEM enrichment summer program for high school students. We hope that our model and methods will allow other researchers to create lightweight and accurate models as more COVID-19 X-ray scans become available.

Evaluation of a step-by-step approach to frozen section diagnosis in ovarian masses

Background: Ovarian cancer forms a significant proportion of cancer-related mortality in females. It is often detected late due to non-specific clinical presentation. Radiology and tumor markers may indicate an ovarian mass. However, exact diagnosis requires pathological evaluation, which may not be possible before surgery. Intraoperative frozen section (FS) is, therefore, an important modality for the diagnosis of ovarian masses. Aims and Objectives: This study was conducted to study step-by-step approach along with diagnostic utility and accuracy of intraoperative FS in diagnosis of ovarian masses. Materials and Methods: Retrospective comparative analysis was done to determine the diagnostic accuracy of FS as compared to routine histopathology in the pathology department of a tertiary care hospital. Diagnostic categorization was done into benign, borderline, and malignant. Overall accuracy, sensitivity, and specificity of FS technique were calculated. Results: Out of 51 cases, FS analysis yielded accurate diagnosis in 94.1% of ovarian masses. Intraoperative FS had a sensitivity of 94.7%, specificity of 96.9%, 3.1% false-positive rate, and 5.3% false-negative rate in malignant tumors. In benign lesions, FS had 91.7% sensitivity and 100% specificity. FS had 75% sensitivity and 96.4% specificity in cases of borderline tumors. Conclusion: FS is a fairly accurate technique for intraoperative evaluation of ovarian masses. It can help in deciding the extent of surgery. It distinguishes benign and malignant tumors in most cases with high sensitivity and specificity. A methodical approach is useful in determining accurate diagnosis on FS diagnosis.

Hidden biases in germline structural variant detection

Background Genomic structural variations (SV) are important determinants of genotypic and phenotypic changes in many organisms. However, the detection of SV from next-generation sequencing data remains challenging. Results In this study, DNA from a Chinese family quartet is sequenced at three different sequencing centers in triplicate. A total of 288 derivative data sets are generated utilizing different analysis pipelines and compared to identify sources of analytical variability. Mapping methods provide the major contribution to variability, followed by sequencing centers and replicates. Interestingly, SV supported by only one center or replicate often represent true positives with 47.02% and 45.44% overlapping the long-read SV call set, respectively. This is consistent with an overall higher false negative rate for SV calling in centers and replicates compared to mappers (15.72%). Finally, we observe that the SV calling variability also persists in a genotyping approach, indicating the impact of the underlying sequencing and preparation approaches. Conclusions This study provides the first detailed insights into the sources of variability in SV identification from next-generation sequencing and highlights remaining challenges in SV calling for large cohorts. We further give recommendations on how to reduce SV calling variability and the choice of alignment methodology.

Urgent care study of the LumiraDx SARS-CoV-2 Ag Test for rapid diagnosis of COVID-19

Background Testing individuals suspected of severe acute respiratory syndrome–like coronavirus 2 (SARS-CoV-2) infection is essential to reduce the spread of disease. The purpose of this retrospective study was to determine the false negativity rate of the LumiraDx SARS-CoV-2 Ag Test when utilized for testing individuals suspected of SARS-CoV-2 infection. Methods Concurrent swab samples were collected from patients suspected of SARS-CoV-2 infection by their healthcare provider within two different urgent care centers located in Easton, MA, USA and East Bridgewater, MA, USA. One swab was tested using the LumiraDx SARS-CoV-2 Ag Test. Negative results in patients considered at moderate to high risk of SARS-CoV-2 infection were confirmed at a regional reference laboratory by polymerase chain reaction (PCR) using the additional swab sample. The data included in this study was collected retrospectively as an analysis of routine clinical practice. Results From October 19, 2020 to January 3, 2021, a total of 2241 tests were performed using the LumiraDx SARS-CoV-2 Ag Test, with 549 (24.5%) testing positive and 1692 (75.5%) testing negative. A subset (800) of the samples rendering a negative LumiraDx SARS-CoV-2 Ag Test was also tested using a PCR-based test for SARS-CoV-2. Of this subset, 770 (96.3%) tested negative, and 30 (3.8%) tested positive. Negative results obtained with the LumiraDx SARS-CoV-2 Ag test demonstrated 96.3% agreement with PCR-based tests (CI 95%, 94.7–97.4%). A cycle threshold (CT) was available for 17 of the 30 specimens that yielded discordant results, with an average CT value of 31.2, an SD of 3.0, and a range of 25.2–36.3. CT was > 30.0 in 11/17 specimens (64.7%). Conclusions This study demonstrates that the LumiraDx SARS-CoV-2 Ag Test had a low false-negative rate of 3.8% when used in a community-based setting.