A novel cervical image recognition framework based on feature cognitive enhancement for improved performance.

Cervical cancer ranks as the fourth most prevalent cancer among women worldwide. Early diagnosis facilitates timely intervention and treatment. Traditional colposcopy is a widely employed technique for evaluating cervical lesions. Nevertheless, the growing number of patients with cervical cancer, and the rising workload of doctors may lead to misdiagnosis and underdiagnosis. Therefore, many researchers use deep learning to make paramedical diagnoses of cervical disease. However, the current neural network models for cervical image recognition lead to poor diagnostic results due to limitations in feature extraction. Consequently, this study proposes a novel model for classifying cervical images into normal, cervical intraepithelial neoplasia, and cancerous categories. The model proposes three new modules, namely a Feature Cognitive Screening Module, Multi-scale Feature Classification Module, and Overlapping Sampling Module (FSMO), which can realize the extraction of global features and local feature areas, multi-scale feature fusion classification and short-range interactions in the images, and enhance the edge capturing ability of the model and the proficiency of solving complex problems, thereby elevating prediction accuracy. The results showed that the recognition accuracy of FSMO in the self-constructed cervical image dataset was 91.88%, a precision of 92.91%, a recall of 91.92%, and an F1-Score of 91.99%; and the accuracy is 97.5% under the kaggle dataset. The performance surpasses other advanced models. Consequently, this model holds significant potential for rapid auxiliary diagnosis in cervical imaging, contributing to the early detection and treatment of cervical cancer.