Machine Learning Assisted Cervical Cancer Detection.

Area of Science:

• Oncology
• Medical Informatics
• Machine Learning

Background:

• Cervical cancer is a leading cause of cancer death in women worldwide, strongly linked to human papillomavirus (HPV).
• Limited access to screening in developing nations increases patient risk due to cost, awareness, and healthcare access barriers.
• Identifying individual patient risk factors is crucial for proactive intervention and reducing the global burden of cervical cancer.

Purpose of the Study:

• To propose and evaluate CervDetect, a novel machine learning approach for assessing cervical cancer risk factors.
• To leverage advanced algorithms for improved accuracy in cervical cancer detection and risk prediction.
• To address disparities in cervical cancer screening by offering a potentially more accessible risk assessment tool.

Main Methods:

• Utilized Pearson correlation for data pre-processing, analyzing relationships between input variables and the outcome.
• Employed the Random Forest (RF) algorithm for effective feature selection, identifying key risk indicators.
• Developed a hybrid model combining RF and shallow neural networks for robust cervical cancer detection.

Main Results:

• CervDetect demonstrated high predictive accuracy for cervical cancer, achieving 93.6%.
• The model reported a Mean Squared Error (MSE) of 0.07111, indicating precise predictions.
• Achieved a False Positive Rate (FPR) of 6.4% and a False Negative Rate (FNR) of 100%.

Conclusions:

• CervDetect accurately predicts cervical cancer risk, outperforming existing state-of-the-art methods.
• The hybrid RF and neural network approach offers a promising tool for early detection and risk assessment.
• Further development and implementation of CervDetect could significantly enhance cervical cancer screening accessibility and effectiveness globally.