An optimized deep learning model using Mutation-based Atom Search Optimization algorithm for cervical cancer detection

Abstract

The cervical cancer patient’s death rate can be minimized by accurate and early detection of cervical cancer (CC). One of the popular techniques called the Pap test or Pap smear is widely used for the early detection of CC. In the instance of CC detection, the manual analysis took more time. Existing approaches had a number of drawbacks, including low accuracy, increased computational complexity, higher feature dimensionality, poor reliability, and increased time consumption due to poor hyperparameter optimization. In this paper, we proposed MASO-optimized DenseNet 121 architecture for the early detection of cervical cancer. At first, different kinds of augmentation techniques such as horizontal flip, vertical flip, zooming, shearing, height shift, width shift, rotation, and brightness increase the number of training samples. The Mutation-based Atom Search Optimization (MASO) algorithm is established to optimize the hyperparameters in DenseNet 121 architecture such as the number of neurons in the dense layer, learning rate value, and the batch sizes. The proposed method effectively optimizes the hyperparameters inherent in the DenseNet 121 architecture, resulting in improved classification results while reducing computational complexity and data overfitting. Different kinds of performance metrics such as accuracy, specificity, sensitivity, precisions, recall, F-score, and confusion matrix evaluate the performance of MASO-optimized DenseNet 121 architecture for CC detection. A single normal class with three abnormal classes, namely Carcinoma, Light dysplastic, and Sever dysplastic, was selected from the Hervel dataset for experimental investigation. The proposed MASO-optimized DenseNet 121 architecture achieves 98.38% accuracy, 98.5% specificity, 98.83% sensitivity, 98.58% precision, 99.3% recall, and 98.25% F-score values than other existing methods.