Publication Date: 2021/08/01
Abstract: Pneumonia is a lung infection mainly caused by microbes where lungs become inflamed and tiny air sacs (alveoli) get filled with fluids causing difficulty in breathing. As stated by the World Health Organization (WHO), pneumonia is the single largest infectious cause of death in children worldwide accounting for 15% of all deaths of children under five years old. While young and healthy adults have low risk, older people have a greater chance of having pneumonia and are much more likely to die from it. The most convenient way to diagnose pneumonia is through chest x-rays. Deep Learning has shown some tremendous results in medical image analysis in recent times. Convolution Neural Networks (CNNs) are widely used in various classification problems starting from handwritten digit recognition to self-driving cars. However, training a CNN model from scratch could be a tedious task as it requires a huge labeled training data, extensive computational resources for training the model, and it often leads to overfitting and convergence issues. Hence, a convenient alternative for traditional CNN is to fine-tune a pre-trained CNN that has been trained using a large dataset. In this paper, we present the performance analysis of transfer learning and fine-tuning CNN for classifying pneumonia among the chest x-ray samples. Our proposed Fine-Tuned CNN model classifies pneumonia infected chest x-rays into 3 categories bacterial, normal, and viral achieves an accuracy of 83.33% which is comparable to the performance of human radiologists.
Keywords: Convolution Neural Network, Fine-Tuning, Transfer Learning, Chest X-rays, Medical Imaging.
DOI: No DOI Available
PDF: https://ijirst.demo4.arinfotech.co/assets/upload/files/IJISRT21JUL698.pdf
REFERENCES