Lung Disease Classification in CXR Images Using Hybrid Inception-ResNet-v2 Model and Edge Computing

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Science

Background:

• Chest X-ray (CXR) is a vital diagnostic tool, but accurate interpretation is challenging for radiologists.
• A global shortage of trained radiologists limits access to timely diagnoses.
• Automating CXR analysis can improve diagnostic efficiency and accessibility.

Purpose of the Study:

• To evaluate machine learning (ML), deep learning (DL), and transfer learning (TL) models for CXR disease classification.
• To develop an automated system for detecting pneumonia, COVID-19, and normal cases in CXR images.
• To compare the performance of various AI models on an open-source CXR dataset.

Main Methods:

• Utilized a hybrid Inception-ResNet-v2 transfer learning model with data augmentation and image enhancement.
• Employed synthetic minority over-sampling technique (SMOTE) and weighted class balancing to address class imbalance.
• Deployed the best-performing model in an edge environment using Amazon IoT Core for automated detection.

Main Results:

• The hybrid Inception-ResNet-v2 model achieved a high average accuracy of 98.66%.
• Other transfer learning models showed varying accuracies: SqueezeNet (97.33%), VGG19 (91.66%), ResNet50 (90.33%), MobileNetV2 (76.00%).
• A deep learning model trained from scratch reached 92.43% accuracy, while ML models (SVM+LBP, DT+HOG) achieved 87.98% and 86.87% respectively.

Conclusions:

• The proposed hybrid transfer learning approach significantly outperforms other methods for CXR disease classification.
• Automated CXR analysis using AI can effectively aid in diagnosing common respiratory conditions.
• The developed edge-deployed system offers a scalable solution to support healthcare professionals and mitigate radiologist shortages.