Deep learning-based classification of pleural malignancy using medical thoracoscopic images

Area of Science:

• Medical imaging
• Artificial intelligence
• Oncology

Background:

• Malignant pleural effusion (MPE) is a common complication of advanced lung cancer.
• Accurate diagnosis of MPE is crucial for effective treatment planning.
• Medical thoracoscopy (MT) offers direct visualization and biopsy but relies on operator expertise and pathological confirmation.

Purpose of the Study:

• To develop and evaluate a deep learning model for classifying pleural malignancy using thoracoscopic images.
• To assess the diagnostic performance of the AI model in identifying cancerous pleural effusions.

Main Methods:

• A convolutional neural network (InceptionV3) was trained using 4,932 thoracoscopic images from 426 patients.
• Transfer learning and online augmentation were employed during model training.
• Performance was assessed using accuracy, precision, recall, F1 score, and ROC-AUC, with Grad-CAM for visualization.

Main Results:

• The deep learning model achieved 81.7% accuracy, 82.4% precision, 86.6% recall, and an 84.6% F1 score.
• The area under the ROC curve (AUC) was 0.90, indicating excellent discriminative ability.
• Grad-CAM confirmed the model focused on relevant pleural abnormalities, even in subtle cases.

Conclusions:

• This is the first deep learning model for classifying pleural malignancy from MT images.
• The model demonstrates strong diagnostic performance, potentially aiding real-time clinical decisions during thoracoscopy.
• The AI tool can help identify malignant targets for biopsy or pleurodesis, improving patient management.