Deep learning-based image evaluation for cervical precancer screening with a smartphone targeting low resource...

Area of Science:

• Oncology
• Medical Imaging
• Artificial Intelligence

Background:

• Cervical cancer is a leading cause of death for women globally.
• Early detection of cervical precancer, such as high-grade Cervical Intraepithelial Neoplasia (CIN), is crucial for improving survival rates.
• Current screening methods like Visual Inspection with Acetic acid (VIA) have limitations in low-resource settings due to subjective interpretation.

Purpose of the Study:

• To assess the feasibility of a mobile phone-based cervical cancer screening platform.
• To evaluate the performance of a deep learning algorithm (Automated Visual Evaluation - AVE) and an image quality algorithm on low-end smartphones.
• To determine the accuracy and speed of these components for cervical precancer detection.

Main Methods:

• Refactoring the core AVE algorithm into a new deep learning detection framework for smartphone compatibility.
• Developing an image quality algorithm to localize the cervix and assess image quality.
• Assessing both accuracy and speed of the AVE algorithm and image quality algorithm on a low-end smartphone (Samsung J8).

Main Results:

• The refactored AVE algorithm achieved equivalent accuracy and ran in approximately 30 seconds on a low-end smartphone.
• The developed image quality algorithm localized the cervix and assessed image quality in approximately 1 second, with an Area Under the ROC Curve (AUC) of 0.95.
• These results demonstrate the potential for a functional mobile phone-based cervical cancer screening system.

Conclusions:

• Mobile phone-based cervical cancer screening using the AVE algorithm and image quality assessment is feasible.
• The developed algorithms are accurate and efficient enough to operate on low-end smartphones, enhancing accessibility.
• Field validation is currently underway to further confirm the platform's effectiveness in real-world settings.