Diagnostic performance of artificial intelligence models for pulmonary nodule classification: a multi-model evaluation

Category	Detail
Condition	Pulmonary nodules and lung cancer
Key Mechanisms	AI models detect, measure, and classify pulmonary nodules on CT scans to predict malignancy risk
Target Population	Patients aged >18 years with pulmonary nodules sized 4-30 mm undergoing CT-guided biopsy or surgical resection
Care Setting	Tertiary care hospital radiology and pathology departments

Lung cancer mortality is high due to late-stage diagnosis; early detection of malignant pulmonary nodules improves prognosis.
High-resolution CT increases nodule detection but reduces specificity, increasing the need for accurate classification tools.
Three commercial AI models were evaluated for diagnostic accuracy against histopathology, showing potential to automate nodule classification.

Use high-resolution thoracic CT scans with slice thickness ≤4 mm to detect pulmonary nodules sized 4-30 mm.
Confirm nodule presence and size by experienced radiologists before AI assessment.
Apply AI models to classify nodules as benign, intermediate, or malignant based on proprietary malignancy risk scores.

Integrate AI malignancy risk predictions to support clinical decision-making for follow-up, biopsy, or surgical resection.
Consider nodule size and histopathological subgrouping when interpreting AI results, especially for nodules 5-8 mm.
Recognize AI models are primarily trained on primary lung cancers; metastases may require additional clinical correlation.

Monitor changes in pulmonary nodules via follow-up CT scans, particularly for nodules classified as intermediate risk by AI.
Evaluate AI model performance continuously against histopathological outcomes to ensure diagnostic accuracy.

Limited generalizability and transparency of AI decision-making may impact clinical adoption.
AI models may have variable thresholds and scoring systems, requiring cautious interpretation.
Potential for false positives and negatives necessitates radiologist oversight and correlation with clinical data.

Adults with pulmonary nodules detected on thoracic CT scans undergoing biopsy or resection

AI models provide malignancy risk stratification to guide clinical management but require validation against histopathology and clinical context.

Ensure CT scans meet technical criteria (slice thickness ≤4 mm) for optimal AI model performance.
Use experienced radiologists to confirm nodule detection prior to AI analysis.
Interpret AI malignancy risk scores within the context of nodule size, morphology, and clinical history.
Maintain multidisciplinary collaboration between radiologists, pathologists, and clinicians when integrating AI results.
Recognize limitations of AI models, including restricted nodule size ranges and training biases toward primary lung cancers.

Clinical Scorecard: Evaluation of Artificial Intelligence Models in Classifying Pulmonary Nodules: A Comprehensive Diagnostic Assessment