A lightweight deep learning method to identify different types of cervical cancer

Area of Science:

• Oncology
• Artificial Intelligence
• Medical Imaging

Background:

• Cervical cancer, often stemming from Human Papillomavirus (HPV) infection, is a significant global health concern for women.
• Early detection of cervical cancer, originating from cervical intraepithelial neoplasia (CIN), is crucial for effective treatment and improved patient outcomes.

Purpose of the Study:

• To develop and evaluate a novel, lightweight deep learning model (CCanNet) for efficient and accurate cervical cancer detection.
• To compare the performance of CCanNet against established transfer learning and transformer models using a public dataset.

Main Methods:

• A novel deep learning architecture, CCanNet, was designed, integrating squeeze blocks, residual blocks, and skip layer connections.
• Comparative analysis was performed using the SipakMed dataset, evaluating CCanNet against models like VGG19, MobileNetV2, ConvNeXT, and Swin Transformer.
• Model performance was assessed using metrics including accuracy, precision, recall, and F1 score, with Explainable AI (XAI) used for result validation.

Main Results:

• CCanNet achieved a high accuracy of 98.53%, surpassing all compared state-of-the-art models.
• The proposed CCanNet model demonstrated the lowest parameter count (1,274,663) among the evaluated models, indicating high efficiency.
• Comprehensive evaluation metrics confirmed the superior performance and reliability of CCanNet in cervical cancer detection.

Conclusions:

• The developed CCanNet model offers a highly accurate and computationally efficient solution for cervical cancer detection.
• CCanNet's performance suggests its potential as a valuable tool in clinical settings for early diagnosis and management of cervical cancer.
• The integration of XAI ensures the trustworthiness and interpretability of CCanNet's diagnostic capabilities.