CheXNet and feature pyramid network: a fusion deep learning architecture for multilabel chest X-Ray clinical...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer-Aided Diagnosis

Background:

• Multilabel X-ray image analysis is crucial for clinical diagnosis, capturing pathology co-occurrence and interdependency.
• Accurately diagnosing multiple diseases in a single X-ray is challenging due to complex, multi-level features.
• Existing deep learning models aim to enhance classification performance and multi-probability disease detection.

Purpose of the Study:

• To develop an accurate and fast inference system for diagnosing multiple thoracic diseases from chest X-rays.
• To improve the classification and discrimination of multiple diseases in a single X-ray image.
• To enhance clinical diagnosis support systems with advanced AI capabilities.

Main Methods:

• A novel fusion architecture combining CheXNet and Feature Pyramid Network (FPN) was designed.
• The architecture extracts features at different spatial resolutions, capturing both low-level and high-level semantic information.
• The model was evaluated on the NIH ChestXray14 dataset using Area Under the Curve (AUC) and accuracy metrics.

Main Results:

• The proposed method achieved an average AUC of 0.846 and an accuracy of 0.914.
• The architecture demonstrated superior performance compared to other state-of-the-art approaches.
• The model achieved a rapid inference time of 0.013 seconds per image, outperforming existing methods.

Conclusions:

• The fusion architecture shows significant promise for multilabel disease classification in chest X-rays.
• The developed system offers potential applications for quick and accurate clinical diagnosis.
• The study highlights the effectiveness of integrating CheXNet and FPN for enhanced medical image analysis.