Machine Learning: Using Xception, a Deep Convolutional Neural Network Architecture, to Implement Pectus Excavatum...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiology

Background:

• Pectus excavatum (PE) is a chest wall deformity impacting cardiopulmonary function.
• Frontal chest radiography alone is insufficient for PE diagnosis by radiologists.
• Current diagnostic methods often require additional imaging views or computed tomography.

Purpose of the Study:

• To develop and evaluate a deep learning model for pectus excavatum screening using frontal chest X-rays.
• To assess the feasibility of using artificial intelligence for early PE detection in routine clinical practice.
• To improve the accessibility and efficiency of PE screening.

Main Methods:

• A convolutional neural network (CNN) algorithm, specifically Xception, was trained on a database of chest X-rays.
• The dataset comprised posteroanterior-view chest images from patients with and without PE.
• The CNN model was trained using 80% of the data and validated on the remaining 20%.

Main Results:

• The AI model achieved high diagnostic performance, with an area under the receiver operating characteristic curve ranging from 0.976 to 1.
• The model demonstrated a test accuracy of 0.989.
• Excellent sensitivity (96.66%) and specificity (96.64%) were recorded for PE detection.

Conclusions:

• This study is the first to demonstrate the efficacy of a CNN for diagnosing pectus excavatum from frontal chest X-rays.
• The developed AI tool offers a convenient and effective method for screening PE candidates.
• This approach enhances diagnostic capabilities beyond human interpretation of standard chest X-rays.