A Deep Network Incorporating Depthwise Separable Convolutions for Pathological Diagnosis of Chest X-Ray Images: A...

Area of Science:

• Medical Imaging Analysis
• Deep Learning in Radiology
• Artificial Intelligence in Healthcare

Background:

• Pathological diagnosis from chest X-ray images presents significant challenges.
• Accurate and efficient interpretation of radiographic findings is crucial for patient care.
• Existing deep learning models require optimization for improved performance.

Purpose of the Study:

• To develop and evaluate a novel deep learning network for enhanced chest X-ray pathology detection.
• To improve the accuracy and efficiency of automated disease identification in radiographic images.
• To address limitations in current chest X-ray analysis techniques.

Main Methods:

• Proposed TDCheXNet, a network fusing two depthwise separable convolutions for pathology detection.
• Modified CheXNet by embedding depthwise separable convolutions for down-sampling, removing the transformation layer.
• Enhanced the initial convolution layer with an additional depth-separable convolution to capture more pathological information.

Main Results:

• TDCheXNet achieved an 82.8% area under the receiver operating characteristic curve (AUROC) on the ChestX-ray14 dataset.
• The network demonstrated a detection speed of 178.794 ms per image.
• Compared to the original model, TDCheXNet improved average AUROC by 0.5% and increased inference speed by 14.946 ms.

Conclusions:

• The proposed TDCheXNet network offers superior performance for chest X-ray pathology detection.
• Experimental results confirm the network's effectiveness on a large public dataset.
• The study highlights the potential of depthwise separable convolutions for medical image analysis.