A Hybrid Approach for Differentiating Fibrotic Hypersensitivity Pneumonitis and Idiopathic Pulmonary Fibrosis: Deep...

Area of Science:

• Pulmonary Medicine
• Artificial Intelligence in Radiology
• Medical Imaging Analysis

Background:

• Idiopathic pulmonary fibrosis (IPF) and fibrotic hypersensitivity pneumonitis (HP) present similar radiological patterns on high-resolution computed tomography (HRCT).
• Accurate differentiation is crucial for appropriate patient management, yet can be diagnostically challenging, especially in specific patient demographics (e.g., males over 60, non-smokers).

Purpose of the Study:

• To evaluate the efficacy of deep learning-based feature extraction combined with machine learning classification for differentiating IPF from fibrotic HP.
• To develop a robust, non-invasive decision-support tool for challenging diagnostic scenarios in interstitial lung diseases.

Main Methods:

• Retrospective study of 87 patients (52 IPF, 35 HP) with HRCT images.
• Utilized a pre-trained AlexNet for deep feature extraction (fc8 layer) without fine-tuning.
• Implemented rigorous 10-repeated fivefold patient-level cross-validation with hierarchical random undersampling to prevent data leakage and class imbalance.
• Classified extracted features using Support Vector Machine (SVM) and K-Nearest Neighbor (KNN) algorithms.

Main Results:

• The SVM classifier achieved an average accuracy of 83.03%, sensitivity of 81.74%, specificity of 84.29%, F1-score of 82.32%, and an AUC of 91.93%.
• The KNN classifier achieved an average accuracy of 80.56% and an AUC of 87.66%.
• The methodology demonstrated strong generalization capability, overcoming common AI flaws like data leakage and bias.

Conclusions:

• Deep learning feature extraction combined with machine learning classification offers a highly accurate and reliable method for distinguishing IPF from fibrotic HP.
• The proposed framework serves as a valuable non-invasive decision-support tool, particularly in complex cases where radiological differentiation is difficult.