Quantitative Screening of Cervical Cancers for Low-Resource Settings: Pilot Study of Smartphone-Based Endoscopic...

Area of Science:

• Medical Imaging
• Machine Learning
• Oncology

Background:

• Cervical cancer (CC) disproportionately affects low- and middle-income countries, where traditional screening methods are challenging.
• Visual inspection with acetic acid (VIA) is a promoted screening method, but its accuracy can be limited by variability.
• Smartphone-based imaging offers a cost-effective and accessible alternative for medical evaluations.

Purpose of the Study:

• To develop and evaluate a novel smartphone-based endoscopic VIA system for cervical cancer screening.
• To implement and assess machine learning algorithms for analyzing VIA images to improve diagnostic accuracy.
• To compare the performance of the developed system against the gold standard (histopathology) and physician interpretations.

Main Methods:

• A smartphone-based endoscope system was developed for visual inspection with acetic acid (VIA) screening.
• Endoscopic VIA images from 20 patients (5 healthy, 15 with cervical intraepithelial neoplasia) were acquired.
• Image processing techniques extracted features, and machine learning classifiers (KNN, SVM, DT) were compared for accuracy, with KNN selected as the best performing model.

Main Results:

• Physician evaluation of the system yielded an average accuracy of 78%.
• The k-nearest neighbors (KNN) machine learning model achieved 78.3% accuracy in cross-validation and 80.8% on an unprovided dataset.
• The KNN model outperformed average physician assessments in identifying abnormal cervical tissues.

Conclusions:

• Smartphone-based endoscopic VIA, coupled with machine learning, demonstrates significant potential as a cervical cancer screening tool.
• The developed system offers a promising, accurate, and practical solution for low-resource settings.
• Further validation of this technology can enhance early detection and management of cervical cancer globally.