Pneumonia classification using quaternion deep learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Science

Background:

• Pneumonia is a lung infection causing respiratory issues, diagnosed via Chest X-rays.
• Computer-Aided Detection (CAD) tools enhance radiologist accuracy in identifying pneumonia.
• Convolutional Neural Networks (CNNs) excel at image object detection.

Purpose of the Study:

• To evaluate the effectiveness of a Quaternion Convolutional Neural Network (QCNN) for pneumonia classification in Chest X-rays.
• To compare the performance of QCNN against traditional CNN architectures for pneumonia detection.

Main Methods:

• A Quaternion Residual Network (QCNN) was trained on a large, public Chest X-Ray dataset.
• The QCNN processes RGB channels as a single unit to extract enhanced features.
• Performance was evaluated using classification accuracy and F-score, and compared to real-valued networks.

Main Results:

• The QCNN achieved a classification accuracy of 93.75% and an F-score of 0.94.
• The Quaternion Residual Network demonstrated higher classification accuracy compared to a real-valued Residual Network.
• QCNNs showed improved feature extraction capabilities for medical image classification.

Conclusions:

• Quaternion Convolutional Neural Networks offer a promising advancement for automated pneumonia detection in Chest X-rays.
• QCNNs provide superior performance over traditional CNNs, aiding in more accurate and efficient diagnosis.
• This approach has the potential to significantly assist radiologists in clinical settings.