A hybrid modeling approach with interpretability in Cervical Cell Classification Using a CNN-ML Synergistic Approach

This study presents research aimed at improving the robustness and interpretability of a Convolutional Neural Network (CNN) model for classifying cervical cell images to detect potential health issues. An affordable and widely utilized early screening technique for cervical cancer, the Pap smear, is frequently constrained by its labor-intensive nature and vulnerability to errors, which substantially diminish its overall effectiveness. This study seeks to address this limitation by developing an automated cancer cell classification approach that integrates features extracted from CNNs with traditional machine learning algorithms, while employing Grad-CAM to enhance interpretability. Applied to the SIPaKMeD dataset, the model achieved 93% accuracy, with the ResNet152V2 and Linear Regression models performing best. Grad-CAM visualizations improved trust in AI-driven cervical cancer classification. These findings demonstrate the potential of AI in improving cervical cancer diagnostics.