AI4COVID: Artificial Intelligence Enabled Abnormality Detection for Chest X-Rays in COVID and Non-COVID Patients

Project Title

AI4COVID: Artificial Intelligence Enabled Abnormality Detection for Chest X-Rays in COVID and Non-COVID Patients

Project Area of Specialization Artificial IntelligenceProject Summary

The COVID-19 pandemic has affected global health services and has a devastating impact on general public health and well being. One of the key factors in the fight against global pandemic is a timely and accurate diagnosis of chest X-rays for the screening of COVID-19 patients. A correct interpretation of Chest X-rays is crucial for correct diagnosis. The situation has been exacerbated by spikes in COVID-19 cases and the corresponding pressure on health systems with limited resources, especially in third world countries like Pakistan. Artificial Intelligence (AI) based diagnosis and classification of lung diseases, such as COVID-19 can enable quick and accurate diagnosis. AI-enabled models can relieve the work pressure from radiologists under a high volume of COVID-19 cases and can enable processing of large amounts of X-ray images in a short span of time. Furthermore, AI-enabled diagnosis of COVID-19 and Non-COVID related lung diseases can augment expert diagnosis with a second opinion. The system could provide a very useful alternative to the lack of medical experts in rural areas, where radiology technicians can utilise AI-based diagnosis to obtain an expert level opinion. 

We propose AI-based classification and diagnosis of Chest X-ray images for detection of COVID-19 and Non-COVID lung infections. We will design and implement an AI model utilising state-of-the-art deep learning techniques for multi-classification and diagnosis of COVID-19 and 13 categories of Non-COVID lung and thoracic diseases from chest X-rays. We will develop and evaluate AI models based on the latest computer vision techniques and deep learning neural networks used for diagnosis and detection of diseases from images. We utilise open-source data with approximately 1700 COVID chest infections [1], [2] and 18000 Non-COVID infections [3], duly annotated and diagnosed by medical experts. Our software model will be deployable to desktop computers, while hardware portable camera-based systems will be able to diagnose directly by capturing chest X-rays in absence of desktop computers for deployment in rural environments. Our proposed model will ultimately lead to an accurate and quick diagnosis of many critical cases and can potentially help to revolutionise diagnostic practices in Pakistan.

[1] https://github.com/ieee8023/covid-chestxray-dataset

[2] https://www.kaggle.com/tawsifurrahman/covid19-radiography-database

[3] https://vindr.ai/datasets/cxr

Project Objectives

• The main objective of this project is the implementation AI-based deep learning model for medical diagnosis - employing convolutional and deep neural networks, in medical imaging of chest being the key clinical focus area

• We plan to localize and classify 14 types of thoracic abnormalities from chest radiographs (X-rays).                                                          All images were labelled by a panel of experienced radiologists for the presence of 14 critical radiographic findings such as lung infections caused by covid-19, Aortic enlargement, Atelectasis, Calcification, Cardiomegaly, Consolidation, ILD6 - Infiltration, Lung Opacity, Nodule/Mass, Other lesions, Pleural effusion, Pleural thickening, Pneumothorax - Pulmonary fibrosis, No findings.

• We aim to create An automated system that can accurately identify and localize findings on chest radiographs using incredible artificial intelligence algorithms. 

• To create a fine model and construct such algorithms that would help in the detection of errors which might remain unnoticed by the physicians— creating a catalogue of tendering tactics powered by optimization and machine learning.  This project aims at analyzing the recent application of various concepts in computer science to each stage of CAD and CADx.Digitizing histopathology slides produces gigapixel images of around 100,000 ×100,000 pixels, whereas typical CNN image inputs are around 200 ×200 pixels-thus employing GPU’s support. 

•  Jupyter Notebook-Anaconda documentation-The aim is to analyze Computer-Aided Detection(CAD) and (Computer-Aided Diagnosis) CADx systems for detection of 2D and 3D medical image data for multi-class classification and image segmentation. It is necessary to review the various techniques followed in CAD and CADx systems proposed and implemented by ANN and DNN.

• Real-World Application-  Ultimately create an integrated software and hardware integrated system used in clinics and hospitals to increase diagnostic accuracy for physicians

• Research Potential- Collaborate with hospitals like Shaukat Khanum and Agha Khan for research objectives and acquisition of custom data sets. 

 

Project Implementation Method

The major implementation of this project is on Tensorflow which is a rich machine learning library that enables the user to create many neural network layers. The major focus of this library is on training the deep neural networks.

The hardware implementation of the project characterizes the use of JETSON NANO for the execution of image classification and object detection. Generally, this hardware is used to run deep learning neural networks in parallel.

Further execution of the project will involve disease classification, nodule detection, and region segmentation of the thorax cavity. These models can be developed for most conditions for which data can be collected. This will enable the model to respond rapidly in times of crisis—for instance, developing and deploying COVID-19 detection models.

Computer-aided detection systems usually take input images for a series of preprocessing steps. The main purpose of preprocessing is to enhance the quality of the images and make the ROI (region of interest) more obvious. Thus, the quality of the preprocessing has a large influence on the performance of the subsequent procedures. Typical preprocessing techniques include image enhancement, image segmentation and bone suppression for specific applications in chest X-rays. 

Benefits of the Project

Computer-aided detection and diagnosis systems (CADe/CADx) would help reduce the pressure on doctors at metropolitan hospitals and improve diagnostic quality in rural areas. Existing methods of interpreting chest X-ray images classify them into a list of findings. There is currently no specification of their locations on the image which sometimes leads to inexplicable results. A solution for localizing findings on chest X-ray images is needed for providing doctors with the explainable diagnosis. 

This project could act as a valuable second opinion for radiologists. An automated system that could accurately identify and localize findings on chest radiographs would relieve the stress of busy doctors while also providing patients with a more accurate diagnosis. In terms of precision medicine and the pandemic, this project can be used to increase outcome precision and accuracy and to potentially identify patterns within patient data to determine their probability of having a specific disease or illness. The data set is curated by expert radiologists, thus increasing the chance of accurate diagnosis even more. 

Our proposed project can take the strain off the already overstressed and compromised healthcare system in this ongoing pandemic. With machines doing the job, patients can have reduced risk of exposure and dependable results without multiple visits to the doctor. Furthermore, data can be shared in real-time with practitioners such as doctors, medical staff, scientists, research labs all over the world due to AI-based practices. Any doctor anywhere in the world can now access databases and have all the latest insights in seconds as the system will be able to give a diagnosis in seconds, against the traditional practice of lengthy doctor visits and consultations.

It also puts consumers in control of health and well-being. Additionally, it will increase the ability for healthcare professionals to better understand the day-to-day patterns and needs of the people they care for, and with that understanding, they can provide better feedback, guidance and support for staying healthy.

This level of insight can have immense value to the medical profession as it can fully streamline patient care and reduce potential risks by addressing their root causes earlier. Our project's ability to read and analyze vast quantities of information is the key to unlocking the full potential of precision medicine. A14COVID also has excellent use of potential in this global COVID-19 pandemic. Instead of reliance on traditional epidemiological and medical tools, the model can detect the unusual case of COVID pneumonia earlier, saving many lives. 

Apart from real-world applications, the project holds an excellent research value in the field of both Artificial Intelligence and Healthcare and can revolutionise health diagnosis in Pakistan. 

Technical Details of Final Deliverable

The project focuses on key fields of data science and artificial intelligence: computational biomedicine, computer vision, and medical image processing. The medical imaging will be compiled for research in collecting, processing, analyzing, and understanding medical data. We’ll work to build large-scale and high-precision medical imaging solutions based on the latest advancements in artificial intelligence to facilitate effective clinical workflows.

Digitizing histopathology slides produces gigapixel images of around 100,000 ×100,000 pixels, whereas typical deep learning models, such as Convolutional Neural Network image inputs are around 200 ×200 pixels. Radiology modalities such as X-Rays and HRCT of chest render equally massive 3D images, forcing standard CNNs to either work with a set of 2D slices, or adjust their internal structure to process in 3D. 

3D convolutions in CNNs along with image registration when employed will enable better learning from 3D volumes and enable working with time-series images.

Trained on a custom data set curated by expert practitioners the project will use TensorFlow as it’s Deep Learning Framework and Google Collab and Jetson Nano to fulfil its GPU and Hardware requirements. Preprocessing, comparison and fine-tuning of data will ultimately result in obtaining optimal results. 

The final deliverable will be a hardware and software integrated system able to scan thousands of images in seconds and give precision results for not only 14 thoracic cavity abnormalities but also quick recognition of COVID-19 symptoms. 

Final Deliverable of the Project HW/SW integrated systemCore Industry MedicalOther Industries IT Core Technology Artificial Intelligence(AI)Other TechnologiesSustainable Development Goals Good Health and Well-Being for PeopleRequired Resources

Item Name	Type	No. of Units	Per Unit Cost (in Rs)	Total (in Rs)
			Total in (Rs)	70000
Jetson Nano Developer Kit	Equipment	2	25000	50000
Jetson Nano™ Camera	Equipment	2	5000	10000
Travel Expense for Data Collection	Miscellaneous	1	10000	10000