Published Paper Details:

Cervical cancer continues to be a significant worldwide health concern, particularly in areas with poor access to professional screening. The multi-stage, CPU-friendly deep learning pipeline shown in this study is intended to facilitate accessible and comprehensible colposcopy-based diagnosis. Four sequential modules make up the framework's clinically inspired workflow: a surveyor for image quality assessment, a Screener for normal-abnormal differentiation, a Grader for lesion severity classification, and a Reinforcement Learning (RL) Decision Agent for confidence refinement. The fundamental framework for classification jobs is EfficientNet-B3, which provides robust performance while maintaining computational viability for CPU deployment. This method improves reliability, lowers false positives, and facilitates scalable early detection in healthcare settings with limited resources by organizing the system as a hierarchical pipeline and addressing real-world picture variability.