SurvdigitizeR: an algorithm for automated survival curve digitization.

Area of Science:

• Biostatistics
• Medical Informatics

Background:

• Kaplan-Meier (KM) curves are crucial for survival analysis in medical research.
• Manual digitization of KM curves for data extraction is labor-intensive and prone to errors.

Purpose of the Study:

• To develop and validate an automated algorithm for extracting survival probabilities from KM curves.
• To improve the efficiency and accuracy of data extraction from published survival data.

Main Methods:

• The algorithm processes image files (JPG, PNG) using color space conversion and optical character recognition.
• K-medoids clustering is employed to differentiate overlapping curves within figures.
• Performance was validated against manual digitization using simulated data and real-world published KM curves, with Root Mean Squared Error (RMSE) and Bland-Altman analysis.

Main Results:

• Automated digitization demonstrated high accuracy in extracting survival probabilities from simulated KM curves, with an average RMSE of 0.012.
• The algorithm's accuracy was slightly reduced by a higher number of curves per figure and the presence of censoring markers.
• Real-world validation showed strong agreement between automated and manual digitization methods.

Conclusions:

• The developed algorithm automates KM curve digitization, requiring minimal user input and offering accuracy comparable to manual methods.
• This tool streamlines data extraction for decision analytic models and meta-analyses.
• The algorithm is available as an open-source R package and a Shiny application on GitHub.