Advancing Cervical Cancer Prediction: Comparative Analysis of VGGNet, GoogleNet, and DenseNet121 via Deep Learning

• Published in: 2024 Asia Pacific Conference on Innovation in Technology (APCIT) pp. 1 - 4
• Main Authors: Bordoloi, Dibyahash, Joshi, Kireet, Kukreja, Vinay, Sharma, Rishabh
• Format: Conference Proceeding
• Language: English
• Published: IEEE 26.07.2024
• Subjects: Accuracy; Analytical models; Cervical Cancer; Deep learning; DenseNet121; Diagnostic Accuracy; GoogleNet; Magnetic resonance imaging; Medical Imaging; Roads; Sensitivity; Technological innovation; VGGNet
• DOI: 10.1109/APCIT62007.2024.10673598

This research compares the performance of three deep learning models- VGGNet, Googlenet, and Densenet-121 in designing cervical cancer images for diagnostic purposes. This is to help identify the model that is suitable and precise for diagnostic accuracy in clinical settings. A dataset covering digital images of cervical cells which includes normal, pre-cancerous, and cancerous cases, was formulated, inside which the models were tested. The fine-tuning of each model was done with this dataset as inputs, and then a comparative analysis was performed based on accuracy, sensitivity, specificity, and precision. The excessive accuracy of DenseNet121 was 94.7% with a sensitivity of 93.5%, as well as a specificity of 95.9% and a precision of 94.4%. GoogleNet and VGGNet were those that performed highly, although they showed slightly less efficacy than DenseNet121 when applied for this use. As a result of the study being carried out, it turns out that DenseNet121 is more efficient than VGGNet and GoogleNet in terms of detecting cervical cancer where it can therefore be taken to mean that DenseNet121 is thus the best model for the detection and diagnosis of cervical cancer. The results imply a road for deep learning models could be able to provide a supplement to traditional screening techniques and can guide the cost-efficient and non-invasive means to diagnose cervical cancer effectively. It is imperative to do further examination on the use of these models in diverse populations to confirm the results and also integrate them into clinical workflow to assess how they are useful in diagnosis accuracy and patient care.