Optimized hybrid CNN-RNN model with grid search hyperparameter tuning for enhanced diagnostic accuracy in cervical cancer detection

Cervical cancer is among the top ten causes of death among women in the world, and early detection is imperative for effective treatment that may improve outcomes. Traditional methods developed for diagnosing cervical cancer usually fail in low-resource settings. This work developed an optimized hybrid CNN-RNN model, which combined the strengths of CNN spatial feature extraction with RNN temporal analysis for improved cervical cancer image classification. The paper is based on the analysis of the effectiveness of the hybrid model compared to standalone models like CNN, RNN, MLP, and LSTM. Each model was trained and then tested with a labeled dataset containing cervical cancer images, followed by hyperparameter optimization through a grid search. This yielded a very high validation accuracy of 89.64% with a low validation loss of 0.3222, beating the standalone models with significantly lower accuracies: CNN and MLP at ~19%, RNN at 59.28%, and LSTM at 74.28%. The AP scores of the hybrid model were very high across classes, showing that the proposed model would be highly capable of minimizing false positives and negatives. This leads to the conclusion that the CNN-RNN model can provide a trustworthy diagnostic solution that is clinically applicable, especially in settings with limited resources. The high accuracy and the balance of precision-recall present an excellent opportunity for this tool to be used in early cervical cancer detection. Thus, it would support better patient outcomes and could lead to reduced mortality rates.