AI-driven deep CNN approach for multi-label pathology classification using chest X-Rays | JoVE Visualize

AI-driven deep CNN approach for multi-label pathology classification using chest X-Rays

Saleh Albahli¹, Hafiz Tayyab Rauf², Abdulelah Algosaibi³

¹Department of Information Technology, College of Computer Science, Qassim University, Buraydah, Saudi Arabia.

Peerj. Computer Science|May 12, 2021

View abstract on PubMed

Summary

This study introduces synthetic data augmentation with deep learning models to detect 14 chest diseases from X-rays. The approach achieved an average ROC-AUC score of 0.80, improving multi-disease detection accuracy.

Area of Science:

Medical Imaging
Artificial Intelligence
Computer Science

Background:

Artificial intelligence (AI) is crucial for chest X-ray analysis, but detecting multiple diseases simultaneously remains challenging.
Existing multi-disease detection methods suffer from insufficient and imbalanced datasets, limiting their clinical utility.

Purpose of the Study:

To develop an effective method for detecting 14 distinct chest-related diseases using AI.
To address data limitations in multi-class chest X-ray classification through synthetic data augmentation.

Main Methods:

Three deep Convolutional Neural Networks (CNNs) architectures—DenseNet121, InceptionResNetV2, and ResNet152V2—were employed.
Synthetic data augmentation was utilized to overcome data scarcity and imbalance issues.
The models were trained and validated for the classification of 14 chest diseases.

Main Results:

An average Receiver Operating Characteristic Area Under the Curve (ROC-AUC) score of 0.80 was achieved.
The proposed model demonstrated competitive performance compared to previous multi-class classification approaches.
The study highlights the effectiveness of deep neural networks in classifying multiple chest diseases.

Conclusions:

Synthetic data augmentation combined with deep CNNs significantly improves the accuracy of detecting multiple chest diseases.
The developed model offers a promising advancement for AI-driven diagnostic tools in radiology.
This research contributes valuable insights for both the healthcare industry and AI research communities.

Keywords: