Automatic cervical cancer classification using adaptive vision transformer encoder with CNN for medical application

Abstract

Accurate and early cervical cancer screening can reduce the mortality rate of cervical cancer patients. The Pap test, often known as a Pap smear, is one of the frequently used methods for the early diagnosis of cervical cancer. However, manual analysis can be time-consuming. Previous approaches have faced challenges such as low accuracy, increased computing complexity, larger feature dimensionality, poor reliability, and increased time consumption due to subpar hyper-parameter optimization. This paper proposes an automatic cervical cancer classification system using a deep learning algorithm to address these issues. The proposed system consists of three stages: pre-processing, segmentation, and classification. Initially, images are collected and pre-processed through normalization, smoothing, and resizing. The pre-processed images are then passed to the segmentation stage, where an Adaptive Deep Residual Aggregation Network is utilized (ADRAN). After segmentation, the images are classified into seven categories: Carcinoma_in_situ, Light_dysplastic, Moderate_dysplastic, Normal_columnar, Normal_Intermediate, Normal_superficial, and Severe_dysplastic using an Adaptive Vision Transformer Encoder (AVTE) with CNN. To improve the efficiency of the transformer learning network, the hyperparameters of AVTE with CNN are optimized using an Adaptive Cat Swarm Optimization algorithm (ACSO). The efficiency of the presented technique is evaluated based on various metrics, and experimentation is conducted using the Herlev dataset.