Development of Predictive Models in Patients with Epiphora Using Lacrimal Scintigraphy and Machine Learning

Area of Science:

• Ophthalmology
• Medical Informatics
• Machine Learning

Background:

• Epiphora, characterized by excessive tearing, can stem from anatomical or functional obstructions in the lacrimal system.
• Accurate diagnosis is crucial for effective treatment, but traditional methods can be time-consuming and subjective.

Purpose of the Study:

• To develop and evaluate machine learning (ML) and deep learning (DL) predictive models for classifying clinical diagnoses in patients with epiphora.
• To compare the diagnostic performance of these models across different programming languages and computing platforms.

Main Methods:

• 250 patients with epiphora underwent dacryocystography (DCG) and lacrimal scintigraphy (LS).
• Five ML models were developed using Python (TensorFlow), R, and Microsoft Azure Machine Learning Studio (MAMLS), incorporating 27 clinical parameters.
• Two convolutional neural network (CNN) models were created for diagnosing LS images using supervised learning.

Main Results:

• The ML models achieved test accuracies ranging from 73.10% to 81.70% when classifying epiphora diagnoses.
• CNN models demonstrated test accuracies of 72.00% for three-class classification and 77.42% for binary classification.
• Model performance was comparable to that of a nuclear medicine physician.

Conclusions:

• ML-based predictive models are effective tools for classifying clinical diagnoses in patients with epiphora.
• The developed models, utilizing diverse programming languages and platforms, demonstrate diagnostic capabilities similar to experienced clinicians.
• These findings suggest ML and DL hold significant potential for improving the accuracy and efficiency of epiphora diagnosis.