PneumoniaNet

A neural network for COVID-19 detection Blog

Deep neural network for COVID-19 detection in Chest X-rays

Domain 		: Computer Vision, Machine Learning
Sub-Domain	: Deep Learning, Image Recognition
Techniques	: Deep Convolutional Neural Network, Transfer Learning, VGG19
Application	: Image Classification, Medical Imaging, Bio-Medical Imaging

Description

Description

• Identification of COVID-19 pneumonia positive chest X-rays from other non COVID-19 viral pneumonia chest X-rays.
• COVID19 positive images were collected from covid-chestxray-dataset and viral pneumonia images were collected from NIH Chest X-ray dataset.
• 28 COVID-19 chest X-rays and 30 non-COVID-19 viral chest X-ray images were set aside as test data.
• ~150 of the remaining COVID-19 chest X-rays were augmented and the increased their size to ~600.
• Employed transfer learning and fine-tuned the pretrained VGG19 Convolutional Neural Network weights to distinguish COVID-19 positive chest X-rays from other viral chest X-rays.
• Used Tensorflow 2.0 for model training. Incrementally unfroze and tuned all layers in the network.
• Attained a loss (categorical crossentropy) 0. 227 and an accuracy 97% on the test data

Dataset Details

Dataset Name :covid-chestxray-dataset, chest X-ray Original Publication: Identifying Medical Diagnoses and Treatable Diseases by Image-Based Deep Learning

Number of Classes : 2

Tools/ Libraries

Languages	    : Python
Tools/IDE	    : Jupyter. Notebook
Libraries	    : TensorFlow 2.0, VGG19

Model Training

Performance Metrics

Validation Dataset

	precision	recall	f1-score	support
COVID	0.98	0.97	0.97	60
non COVID	1.00	1.00	1.00	553
accuracy			1.00	613
macro avg	0.99	0.98	0.99	613
weighted avg	1.00	1.00	1.00	613

Test Dataset

	precision	recall	f1-score	support
COVID	1.00	0.93	0.96	28
non COVID	0.94	1.00	0.97	30
accuracy			0.97	58
macro avg	0.97	0.96	0.97	58
weighted avg	0.97	0.97	0.97	58

Model and Training Parameters

Parameter	Value
Base Model	VGG19
Optimizer	RMSProp, Stochastic Gradient Descent
Loss Function	Categorical Crossentropy
Learning Rate	0.0001
Batch Size	32
Number of Epochs	Round #15 & #2: 15 epochs, Round#3: 25 epochs

Confusion Matrices

Test Predictions