Introducing papers related to Health applied to Machine Learning

Transcript

1. Sprint 16 Research area: Machine Learning applied to Health By: Alioune THIOUNE
2. Detection of Covid-19 Patients with Convolutional Neural Network Based Features on Multi-class X-ray Chest Images Ali Narin https://arxiv.org/pdf/2012.05525.pdf Research Field •Convolutional Neural Networks Originality •Detection of the presence of Covid-19 by reverse transcription polymerase chain reaction (RT-PCR) tests is of vital importance. Due to its low sensitivity and time consuming, alternative fast and effective methods are needed to be developed. •In this study, a 3-class (Covid- 19, Normal, Viral Pneumonia) detection study has been carried out with the help of SVM using feature maps obtained from the ResNet-50 model, which is a convolutional neural network (CNN) Methodology • The available images are converted from 1024x1024 size to 224x224 size via MATLAB 2020a without pre-processing. All these transformed images are provided as input to the pre- trained ResNet-50 convolutional network model. 1000 features were obtained for each of 2905 data from the layer just before the classification layer. It is submitted to SVM as input with a size of 2905x1000. Findings/Results •The results obtained from this study performed with 3 different functions (SVM Linear, SVM Quadratic, SVM Cubic). •The success of Covid-19 patients detected with the SVM-Quadratic approach is higher than other approaches. At the same time, it is clearly seen that the SVM Quadratic approach gives higher results than other methods in determining healthy individuals. •The overall performance (ACC) results can also be detected with an accuracy of over 99%. •The results obtained are more successful than many studies performed with manual feature extraction methods on Covid19 in the literature. •This study, in which the feature maps obtained from the model are used, is one of the advantageous parts of using the raw data directly and obtaining the features directly without any separate mathematical processing other than the model.
3. Assessing COVID-19 Impacts on College Students via Automated Processing of Free-form Text Ravi Sharma, Sri Divya Pagadala , Pratool Bharti , Sriram Chellappan , Trine Schmidt and Raj Goyal Narin https://arxiv.org/pdf/2012.09369.pdf Research Field • Natural Language Processing Originality •Mental health services are harder to access now during the pandemic due to social distancing and financial hardships of students need to be understood. •Sending periodic surveys or mining social media platforms are not the best options in the context of Covid 19. •This paper discusses an experiment conducted at a US college to assess topics and sentiments of students before and after the pandemic, through processing free-form texts collected via a smartphone app specifically designed for students to express their moods, feelings and opinions. Methodology • Process free-form texts generated by college students (enrolled in a four year US college) via an app specifically designed to assess and improve their mental health. A dataset comprised of more than 9000 textual entries from 1451 students collected over four months (split between pre and post COVID- 19) and established NLP techniques are used. •Determining Topics of Interest Via Contextual and Semantic Categorization • Sentiment Analysis of Topics Pre and Post COVID-19 through NLP models VADER an TextBlob Findings/Results •There is a significant shift in the importance of topics for students before and after “Education” appeared less in students’ posts after COVID-19. The “Health” topic has become the greatest concern while the importance of the “Family” topic is almost the same. There is also an increase in “Housing” and a decrease in “Relationship” topics in the post-COVID-19 timeframe.
4. DEEP NEURAL NETWORKS FOR COVID-19 DETECTION AND DIAGNOSIS USING IMAGES AND ACOUSTIC-BASED TECHNIQUES: A RECENT REVIEW Walid Hariri, Ali Narin https://arxiv.org/pdf/2012.07655.pdf Research Field • Deep Neural Networks Originality • Since the standard testing system of Covid-19 is time- consuming and not available for everyone, alternative early-screening techniques have become an urgent need. • In this paper, the approaches used in the detection of COVID-19 based on deep learning (DL) algorithms, which have been popular in recent years, are comprehensively discussed. Methodology • X-ray images are first pre- processed to remove noise. •Next, deep features are extracted using multiple types of deep models (pre- trained models, generative models, generic neural networks, etc). •Finally, the classification is performed using the obtained features to decide whether the patient is infected by coronavirus or it is another lung disease. Findings/Results •As an alternative to the studies performed on X-ray and CT images, researches on the detection of COVID- 19 are carried out with sound and cough-based acoustic sound analysis. •Traditional ML algorithms are used through the feature extraction part of DL models. This approach appears to improve performance. •It is generally seen that the SVM algorithm is used. •Most of the studies do not use any cross-validation (CV) method, which can decrease the reliability of the results.
5. Effect of Different Batch Size Parameters on Predicting of COVID19 Cases Ali Narin, Ziynet Pamuk https://arxiv.org/pdf/2012.05534.pdf Research Field •Convolutional Neural Networks Originality •In this study, the effect of different batch size (BH=3,10, 20, 30, 40, and 50) parameter values on their performance in detecting COVID19 and other classes was investigated using data belonging to 4 different (Viral Pneumonia, COVID19, Normal, Bacterial Pneumonia) classes. Methodology • Tensorflow-keras are used for the training of the deep learning model. Original images are resized and converted to 224x224. No other pre- processing is applied to the images. •In updating the model weights, the ADAM algorithm, the Epoch number as 30 and the Learning rate as 0.0001 were determined. In addition to the direct transfer of the weights of the ResNet50 model, the model was trained with 3 layers added to the end of the model. Training and test performance of six different batch size (3,10,20,30,40,50) were made. Findings/Results •According to the results, the performance was close on the training and test data. However, it was observed that the steady state in the test data was delayed as the batch size value increased. The highest COVID19 detection was 95.17% for BH = 3, while the overall accuracy value was 97.97% with BH = 20. •Based on the findings, it can be said that the batch size value does not affect the overall performance significantly, but the increase in the batch size value delays obtaining stable results.
6. An Active Learning Method for Diabetic Retinopathy Classification with Uncertainty Quantification Muhammad Ahtazaz Ahsan, Adnan Qayyum, Junaid Qadir and Adeel Razi https://arxiv.org/pdf/2012.13325.pdf Research Field •Bayesian Convolutional Neural Networks Originality •Deep Learning is data-hungry and its training requires extensive computational resources. Another problem with DL is its black-box nature and lack of transparency on its inner working which inhibits causal understanding and reasoning. •In this paper, these challenges are addressed by proposing a hybrid model, which uses a Bayesian convolutional neural network (BCNN) for uncertainty quantification, and an active learning approach for annotating the unlabelled data. Methodology • The proposed model consists of two main components, i.e., Bayesian CNN module and active learning module. Bayesian CNN module is used as feature descriptor by extracting the output of a parametric layer. The active learning module picks an image Xi from the unlabelled data and puts a request to the trained Bayesian CNN module for its label yi, uncertainty ui , and Zi as its respective feature vector. If the uncertainty is less than the threshold value of T, the label is forwarded to the active learning module. Findings/Results •The approach for both binary class classification and multi-class classification was evaluated and competitive results were achieved as compared to the state-of-the- art. The BCNN model for binary class classification achieved an accuracy of 92% and 81%, while the multi-class BCNN model achieved an accuracy of 92%. Moreover, the AL results for the BCNN model for binary class classification are improving state-of- the-art results with an accuracy of 91% and for the case of multi-class classification, AL models for both CNN and BCNN needs to be optimized further.