How AI used to diagnosis the COVID-19

Original article was published on Artificial Intelligence on Medium


How AI used to diagnosis the COVID-19

Introduction:

Many artificial intelligence scientists and researchers have promised that machine learning will change modern medicine with their effective results. There are 1000s of algorithms and models that used to diagnose diseases like cancer, heart disease, infectious disease, and also psychiatric disorders which introduced with the help of AI and Machine learning. Now, this time the algorithms are trained for detecting COVID-19 by analyzing the patterns taken from CT scans and X-ray images of the lungs.

The main target to build this model to predict the patient’s conditions how severe the condition is? According to that, the doctors will provide ventilators and extra cure. These models are really helping out the doctors for treating the patients.

As we know that this COVID-19 has the common symptoms that present in SARS. However, the current situation is challenging to test the patient where they are COVID positive or not, as it doesn’t show the symptoms of viruses. They take time to ensure and carry a possibility of false-negative results. There is a country named Mount Sinai which is the first country for using AI to find out alternative methods to fast and accurate diagnosis of COVID-19 patients by working on imaging and clinical data to analyze the patients with COVID-19.

They have developed an AI model that can detect the COVID-19 in very little time based on how lung disease looks in CT scans including patients’ information, symptoms, age, blood pressure, and possible contact with someone who infected with the COVID-19.

How AI helps to analyze CT images of COVID-19 patients.

This AI-based CT assessment really helps to the physicians with the promising technology. AI can detect the COVID-19 with the deep learning image processing on CT scan images. Gozes et al. published a deep learning model with a sensitivity of 98.2% and specificity of 92.2% based on the thoracic CT algorithm. The main initiative of this model is on focusing on lung infection by analyzing the symptoms that are also common in pneumonia which are also present in COVID-19.

The developers and researchers investigated which symptoms are similar to COVID-19 in order to find out the positive results. They have found many similar symptoms compared to cold, flu, cough, fever, also loss of taste and smell. There are 2/3 people who found corona positive with these symptoms. The result says that the last symptom that is anosmia(loss of taste and smell) is a stronger predictor for finding a COVID-19 positive test.

The AI model used to detect COVID-19:

For training the model, the RT-PCR virology test is used to test whether the report is positive or negative. The scientists and developers have developed three different models using CT scan images and clinical information.

ref: google images.
  1. The very first model is built with deep learning by using CNN(Computational neural network) that uses CT scan’s images to predict COVID-19.
  2. The second model is built with conventional machine learning including SVM(Support vector machine ), Random forest, and multilayer perceptron that uses clinical information only to predict the COVID-19.
  3. The third one is built with CNN that uses radiological and clinical and CT scan images to predict COVID-19.

The CNN model to predict COVID-19.

The model has designed with a supervised task that starts with initializing the weight of the CNN model. The model selects any random image patches of lungs from the trained images and then make labeled patches similar to the trained images. The CNN was trained to classify the images for 1 epoch. This image will classify a local region in the CT scan images to be COVID-19 positive.

The conventional machine learning model.

This model developed with SVM, classification algorithms like random forest and Multi-layered perceptrons based on patient’s age, sex, older health history, current symptoms, blood pressure, white blood cell counts, fever, cough, cold, flu, etc. They have selected the hyperparameters of each classifier by after performing feature engineering on the training set and evaluated the best model such as c and kernel for SVM, the number of estimates in a random forest, and the number of layers and hidden nodes in MLP.

Conclusion:

For this COVID-19 pandemic, every day the researchers and scientists are prioritizing for creating and advanced tools and technologies like Artificial intelligence and machine learning that help to improve the treatment and diagnosis. Also, medical communities are promoting faster detecting models for COVID-19.
AI and ML have the potential to expand the CT scan diagnosis for risk stratification, treatment monitoring, and to provide therapeutic challenges to treat the COVID-19.

Blog references:

  1. https://www.nature.com/articles/s41591-020-0931-3
  2. https://www.genengnews.com/news/covid-19-accurately-diagnosed-by-ai-model/
  3. https://healthitanalytics.com/news/artificial-intelligence-could-speed-covid-19-detection-treatment
  4. https://www.quantib.com/blog/diagnosing-covid-19-using-ai-based-medical-image-analyses

If you want to learn and invest your time in learning Artificial intelligence and machine learning then go for Learnbay. It has also data science courses with the job assistance program that will help you to start your career in the field of AI and machine learning.