Pneumonia Transfer Learning Deep Learning Model from Segmented X-rays.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Science

Background:

• Pneumonia diagnosis from pulmonary radiographs is challenging due to visual similarities with other lung diseases.
• Accurate and rapid pneumonia detection is crucial, especially in resource-limited settings.
• Image processing and deep learning offer promising solutions for automated diagnosis.

Purpose of the Study:

• To develop and evaluate a deep learning model for reliable pneumonia prediction from X-ray images.
• To improve the accuracy and efficiency of pneumonia diagnosis using image segmentation and machine learning.
• To propose an enhanced BoxENet model incorporating transfer learning for superior performance.

Main Methods:

• Utilized a public dataset of 4000 pneumonia and 4000 healthy X-ray images.
• Employed image segmentation with the BoxENet architecture for X-ray analysis.
• Implemented transfer learning using ImgNet and SqueezeNet, combined via a majority fusion model for the improved BoxENet.
• Trained deep learning models for binary and multi-class pneumonia classification.

Main Results:

• The proposed Improved BoxENet model demonstrated superior performance in accuracy, specificity, and sensitivity compared to other models.
• Achieved high Dice scores, indicating effective image segmentation.
• The Improved BoxENet model exhibited faster training and classification speeds.

Conclusions:

• The Improved BoxENet model offers a dependable and efficient method for diagnosing pneumonia from X-ray images.
• Transfer learning integration significantly enhances the model's diagnostic capabilities.
• This approach holds potential for improving pneumonia detection rates and patient outcomes.