Explainable AI Technique in Lung Cancer Detection Using Convolutional Neural Networks

Explainability is essential for clinical adoption: The researchers used SHAP (Shapley Additive Explanations) to visualize which parts of CT images influenced the AI's predictions, addressing a critical barrier to doctor trust. Without seeing why the AI made a decision, clinicians are unlikely to rely on automated screening tools in real medical settings.

Accuracy alone doesn't tell the full story: While ResNet152 achieved the highest accuracy at 97.3%, DenseNet121 was chosen as superior because it better balanced precision (92%), recall (90%), and F1-score (91%). This matters because missing a cancer case (low recall) is more harmful than a false alarm (lower precision), illustrating why responsible AI requires looking beyond single performance metrics.

Addressing data imbalance is a fairness issue: The team used "cost-sensitive learning" to handle unequal numbers of normal, benign, and malignant cases in their dataset. Without this approach, AI systems can develop biases toward the most common category, potentially missing rare but serious conditions—a concrete example of how technical choices impact equitable healthcare.

AI screening could help underserved populations: The authors specifically note potential benefits "in resource-limited settings," suggesting this technology could democratize access to expert-level cancer screening where radiologists are scarce, though implementation would require careful validation in those contexts.