Development of predictive models for lymphedema by using blood tests and therapy data

Area of Science:

• Medical Diagnostics
• Machine Learning in Healthcare
• Lymphatic System Research

Background:

• Lymphedema is characterized by fluid accumulation due to lymphatic system dysfunction.
• Current diagnostic methods are often expensive and may miss early-stage disease.
• Existing machine learning models for lymphedema prediction rely on patient-reported symptoms, which can be inaccurate or absent in early stages.

Purpose of the Study:

• To develop and validate novel machine learning models for early lymphedema detection.
• To investigate the utility of complete blood count (CBC), serum, and therapy data for lymphedema prediction.
• To create a practical tool for clinicians to aid in rapid lymphedema screening.

Main Methods:

• Data from 2137 patients (356 with lymphedema) were collected, including CBC, serum tests, and therapy information.
• Multiple machine learning algorithms (random forest, gradient boosting, decision tree, logistic regression, artificial neural network) were employed.
• Models were trained on 80% of the data and validated on the remaining 20%.

Main Results:

• Random forest models demonstrated strong predictive performance with a balanced accuracy of 87.0% ± 0.7%, sensitivity of 84.3% ± 0.6%, and specificity of 89.1% ± 1.5%.
• The models achieved a precision of 97.4% ± 0.7%, an F1 score of 90.4% ± 0.4%, and an AUC of 0.931 ± 0.007.
• A web application was developed for real-time lymphedema screening by medical practitioners.

Conclusions:

• Machine learning models utilizing CBC, serum, and therapy data offer a promising alternative for early lymphedema detection.
• The developed web application provides a valuable tool for rapid and efficient lymphedema screening.
• This study lays the groundwork for future research into predicting various stages of lymphedema.