Fusing visual and clinical information for lung tissue classification in high-resolution computed tomography.

Area of Science:

• Radiology
• Medical Imaging
• Computer-Aided Diagnosis

Background:

• Accurate classification of lung tissue in high-resolution computed tomography (HRCT) is crucial for diagnosing interstitial lung diseases.
• Integrating clinical data with imaging features can enhance diagnostic performance.

Purpose of the Study:

• To investigate the impact of clinical context on HRCT image-based tissue classification.
• To compare early and late multimedia fusion techniques for combining clinical and visual data.

Main Methods:

• Automatic classification of 2D regions of interest in HRCT axial slices into five lung tissue classes.
• Evaluation of clinical parameter relevance prior to fusion with visual attributes.
• Comparison of early fusion (feature vector concatenation) and late fusion (probability combination via support vector machines).

Main Results:

• The late fusion scheme achieved a maximum of 84% correct predictions for five lung tissue classes.
• This represents a 10% improvement over pure visual-based classification.
• Late fusion demonstrated robustness to the number of clinical parameters used.

Conclusions:

• Late fusion of clinical and visual data significantly enhances HRCT tissue classification accuracy for interstitial lung disease.
• The late fusion approach is suitable for clinical settings, even with missing clinical data.