Pneumonia Diagnosis through pixels -- A Deep Learning Model for detection and classification

Amit Karanth Gurpur,Janani S,Ajeetha B,Brintha Therese A,Rajeswaran Rangasami

2024-04-16

Abstract:Manual identification and classification of pneumonia and COVID-19 infection is a cumbersome process that, if delayed can cause irreversible damage to the patient. We have compiled CT scan images from various sources, namely, from the China Consortium of Chest CT Image Investigation (CC-CCII), the Negin Radiology located at Sari in Iran, an open access COVID-19 repository from Havard dataverse, and Sri Ramachandra University, Chennai, India. The images were preprocessed using various methods such as normalization, sharpening, median filter application, binarizing, and cropping to ensure uniformity while training the models. We present an ensemble classification approach using deep learning and machine learning methods to classify patients with the said diseases. Our ensemble model uses pre-trained networks such as ResNet-18 and ResNet-50 for classification and MobileNetV2 for feature extraction. The features from MobileNetV2 are used by the gradient-boosting classifier for the classification of patients. Using ResNet-18, ResNet-50, and the MobileNetV2 aided gradient boosting classifier, we propose an ensemble model with an accuracy of 98 percent on unseen data.

Image and Video Processing

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to automatically identify and classify pneumonia and COVID - 19 infections in lung CT scan images by using deep - learning models. Manual identification and classification of these diseases is a cumbersome process, and if the diagnosis is delayed, it may cause irreversible damage to patients. Therefore, researchers collected CT scan images from different sources and used pre - trained deep - learning models (such as ResNet - 18, ResNet - 50) and feature extractors (such as MobileNetV2), combined with a gradient - boosting classifier, to build an ensemble classification model to improve the detection and classification accuracy of pneumonia and COVID - 19 infections.

Specifically, the goals of the paper include:
1. **Data collection and pre - processing**: Collect lung CT scan images from multiple data sources and pre - process them through techniques such as normalization, sharpening, median filtering, binarization, and cropping to ensure data consistency when training the model.
2. **Model construction**: Use pre - trained deep - learning models (ResNet - 18, ResNet - 50) for classification and MobileNetV2 for feature extraction. The data after feature extraction is used to train a gradient - boosting classifier.
3. **Ensemble learning**: Through the ensemble learning method, combine the prediction results of multiple models to improve the accuracy and robustness of the overall model.
4. **Performance evaluation**: Evaluate the performance of the model on unseen data, including metrics such as accuracy, precision, recall, and F1 - score.

Finally, the ensemble model proposed in this paper achieved an accuracy of 98% on unseen data, showing its potential in assisting doctors in early and accurate diagnosis.