Quantifying lung fissure integrity using a three-dimensional patch-based convolutional neural network on CT images...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Pulmonology

Background:

• Lung fissure integrity evaluation is crucial for determining emphysema patient eligibility for endobronchial valve (EBV) therapy.
• Accurate assessment of fissure completeness is needed to optimize treatment outcomes.

Purpose of the Study:

• To develop and evaluate a deep learning (DL) approach for segmenting lung fissures on CT images.
• To quantitatively assess fissure integrity in severe emphysema patients using a DL model.

Main Methods:

• A three-dimensional patch-based convolutional neural network (CNN) was trained using 86 CT scans from severe emphysema patients.
• Interlobar regions of interest (ROIs) were annotated to identify fissure presence and absence.
• Fissure integrity was quantified by calculating a fissure integrity score (FIS) from segmented ROIs.

Main Results:

• The DL model achieved a mean absolute percent error between predicted and reference fissure integrity scores of 4.0% (LOF), 6.0% (ROF), and 12.2% (RHF).
• The model demonstrated accurate segmentation and quantitative assessment of fissure integrity across different lung fissures.

Conclusions:

• A DL approach effectively segments lung fissures and quantifies their integrity on CT scans.
• This method shows potential for assisting in the identification of emphysema patients who could benefit from EBV treatment.