Open-source, machine and deep learning-based automated algorithm for gestational age estimation through smartphone...

Area of Science:

• Neonatal Medicine
• Ophthalmology
• Biomedical Engineering
• Artificial Intelligence in Healthcare

Background:

• Accurate gestational age estimation is vital for neonatal care and determining prematurity.
• Current methods may be invasive or inaccessible in resource-limited settings.
• Anterior lens capsule vasculature (ALCV) offers a potential biomarker for gestational age.

Purpose of the Study:

• To develop and validate a fully automated algorithm for estimating gestational age in premature infants.
• To utilize smartphone-based imaging of ALCV for non-invasive gestational age assessment.
• To provide an accessible tool for remote and point-of-care neonatal assessments.

Main Methods:

• A fully automated algorithm employing a fully convolutional network for image segmentation.
• Extraction of ALCV features using a residual neural network architecture.
• Classification of gestational age using a support vector machine trained on extracted features.
• Validation via leave-one-out cross-validation on videos from 124 neonates.

Main Results:

• The algorithm successfully segments usable anterior capsule regions.
• ALCV features are effectively extracted and classified for gestational age estimation.
• Leave-one-out cross-validation demonstrates the algorithm's classification performance.
• The developed software is made open source.

Conclusions:

• A novel, automated algorithm for gestational age estimation using smartphone ALCV imaging has been developed.
• This technology shows promise for accurate and accessible neonatal assessment, particularly in low-income countries.
• The open-source nature of the software facilitates widespread adoption and further research.