A new scheme for knee joint CT image segmentation: A max-flow and watershed method driving semi-automatic CT knee bone segmentation

Knee joint segmentation from CT images is critical for diagnosing and treating arthritis and other knee disorders. Manual segmentation is time-consuming and expert-dependent, highlighting the need for automated, accurate, and efficient methods. This study proposes a scheme for 3D CT knee-joint segmentation that integrates an adaptive weighted continuous max-flow algorithm with a graphical user interface (GUI). Input CT volumes are preprocessed to normalize intensities and improve robustness. Segmentation employs a continuous max-flow formulation with an adaptive weighting function to better capture weak or ill-defined edges in knee CT data. The watershed algorithm is used to resolve adhesions and separate contiguous structures. The approach supports semi-supervised interaction, allowing limited manual guidance when necessary. A graphical user interface (GUI) facilitates data input, interactive refinement, and result export. Performance was assessed on 18 real datasets using precision, sensitivity, and specificity, and was compared against four open-source segmentation tools. Across the 18 datasets, the proposed method achieved higher precision, sensitivity, and specificity than the evaluated open-source tools, demonstrating improved segmentation accuracy and robustness in knee CT images. The proposed semi-automated scheme yields high-precision, efficient knee joint segmentation from 3D CT, reducing manual effort and streamlining the clinical workflow. The improved accuracy and robustness have the potential to enhance diagnostic and treatment planning in orthopedic applications.