Automatic cardiothoracic ratio calculation based on lung fields abstracted from chest X-ray images without heart...

Area of Science:

• Medical Imaging
• Radiology
• Artificial Intelligence in Healthcare

Background:

• The cardiothoracic ratio (CTR) is a key indicator for initial cardiac disease evaluation using postero-anterior chest X-rays (P-A CXR).
• Manual CTR measurement is subjective, time-consuming, and requires expert radiologists due to the difficulty in observing heart borders on P-A CXR images.
• Existing methods face limitations in accurate heart segmentation, leading to potential errors in CTR calculation.

Purpose of the Study:

• To propose a novel, fully automatic method for CTR calculation from P-A CXR images.
• To overcome the challenges of manual heart segmentation and improve the efficiency and accuracy of CTR measurement.
• To develop a practical and feasible tool for the initial evaluation of cardiac diseases.

Main Methods:

• Utilizing convolutional neural networks (CNNs) to abstract lung field mask images from P-A CXR.
• Developing a novel method for localizing the heart's right and left border points based on the 2D projection morphology of lung field masks.
• Implementing graphics-based techniques for precise point localization.

Main Results:

• Achieved mean distance errors of 4.1161 and 3.2116 pixels on static (T1) and dynamic (T2) P-A CXR datasets, respectively.
• Demonstrated mean CTR errors of 0.0208 (T1) and 0.0180 (T2), indicating high accuracy.
• The proposed model showed equivalent performance to the CardioNet model while eliminating the need for heart segmentation.

Conclusions:

• The proposed automatic CTR calculation method is practical and feasible.
• This approach overcomes limitations associated with heart segmentation in manual CTR measurement.
• The method offers a potentially effective tool for the initial assessment of cardiac diseases, reducing time and labor for radiologists.