Stochastic-based learning for image classification in chest X-ray diagnosis.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• Lung diseases, such as pneumonia, pose significant diagnostic challenges.
• Accurate and timely detection of lung conditions from chest X-rays is crucial for effective patient management.

Purpose of the Study:

• To develop and evaluate a stochastic deep learning method for enhanced lung disease detection in chest X-rays.
• To improve diagnostic accuracy and precision for conditions like pneumonia.

Main Methods:

• Utilized a convolutional neural network architecture optimized with dropout regularization and data augmentation.
• Employed stochastic deep learning with stochastic gradient descent, K-Fold cross-validation, and early stopping for model optimization.

Main Results:

• The proposed method demonstrated significant improvements in precision and loss measures across validation folds.
• Achieved a high accuracy of 0.9940 and precision of 0.9960 for pneumonia diagnosis on fold 5.

Conclusions:

• The deep learning strategy offers an effective tool for automated and precise lung disease identification from chest X-rays.
• High accuracy suggests potential for real-world clinical applications, enabling earlier diagnoses and improved patient outcomes.
• Further validation on diverse datasets and integration into clinical decision support systems are recommended for widespread adoption.