Refinement-Cut: User-Guided Segmentation Algorithm for Translational Science

In this contribution, a semi-automatic segmentation algorithm for (medical) image analysis is presented. More precise, the approach belongs to the category of interactive contouring algorithms, which provide real-time feedback of the segmentation result. However, even with interactive real-time contouring approaches there are always cases where the user cannot find a satisfying segmentation, e.g. due to homogeneous appearances between the object and the background, or noise inside the object. For these difficult cases the algorithm still needs additional user support. However, this additional user support should be intuitive and rapid integrated into the segmentation process, without breaking the interactive real-time segmentation feedback. I propose a solution where the user can support the algorithm by an easy and fast placement of one or more seed points to guide the algorithm to a satisfying segmentation result also in difficult cases. These additional seed(s) restrict(s) the calculation of the segmentation for the algorithm, but at the same time, still enable to continue with the interactive real-time feedback segmentation. For a practical and genuine application in translational science, the approach has been tested on medical data from the clinical routine in 2D and 3D.

💡 Research Summary

This paper introduces Refinement‑Cut, an interactive, real‑time segmentation algorithm designed for medical imaging that allows users to guide and correct the segmentation by placing additional seed points without breaking the immediate feedback loop. The method builds upon the Interactive‑Cut framework: a user first places a single seed inside the object of interest, from which a set of radial rays is emitted. Nodes are sampled along each ray, and edge weights are derived from the average gray‑value in the vicinity of the seed. A graph is constructed and a min‑cut is computed to separate foreground from background, providing an instant segmentation result.

In many clinical scenarios the object and background have similar intensities or the object contains internal noise, causing the average gray‑value estimate to be inaccurate. Consequently, a single seed often yields unsatisfactory contours, especially at ambiguous boundaries. Refinement‑Cut addresses this by allowing the user to add one or more seed points directly on the problematic regions. Each additional seed re‑computes the local gray‑value statistics for its ray and neighboring rays, forcing the min‑cut to pass through these points. This local constraint refines the global graph solution, effectively “pulling” the segmentation back to the true boundary.

Performance is evaluated by varying the number of rays and nodes per ray. With 30 rays × 30 nodes (≈900 nodes) the algorithm runs in about 30 ms; with 300 rays × 30 nodes (≈9 000 nodes) it stays under 100 ms, preserving real‑time interaction on a standard laptop (Intel i5‑750, 8 GB RAM, Windows 7). Larger configurations (3 000 rays × 30 nodes ≈90 000 nodes) increase latency to ~130 ms, and 30 000 rays × 30 nodes (≈900 000 nodes) reach ~1 s, which is no longer truly interactive. Thus, practical use recommends keeping the node count below roughly 10 000 for smooth feedback.

The algorithm is demonstrated on four clinical datasets: (1) vertebral body segmentation in MRI, where bright intra‑vertebral regions caused leakage that was corrected by three additional seeds; (2) rectum segmentation in 3 T MRI, where repositioning the initial seed and adding two extra seeds yielded a complete contour; (3) post‑operative abdominal aortic aneurysm CTA, where a stented lumen and surrounding thrombus with an endoleak required multiple seeds to separate lumen, thrombus, and the bright leak region; and (4) 3‑D prostate central gland segmentation in MRI, where a spherical template and a series of seeds produced a volumetric mask that closely matched expert manual delineations.

Quantitatively, a single seed already achieved an average Dice Similarity Coefficient (DSC) of about 80 % across previously published studies. Adding 5–10 seeds raised DSC to >95 % and, in some cases, to 98 %, essentially matching manual expert segmentations. The results demonstrate that the method can handle homogeneous intensity regions, internal noise, and complex shapes with minimal user effort.

Key contributions of the work include: (i) integration of a seed‑based refinement step into a graph‑cut segmentation while preserving real‑time performance; (ii) support for arbitrary 2‑D and 3‑D templates (square, triangle, circle, sphere) allowing flexible adaptation to different anatomical structures; (iii) an intuitive user interaction model that requires only point clicks rather than elaborate strokes or parameter tuning; and (iv) a scalable implementation where computational load can be tuned by adjusting ray and node density.

The authors acknowledge that the current system only accepts point seeds. Future research will explore stroke‑based inputs that trace object boundaries, which could provide richer shape information but may increase graph reconstruction time. Potential solutions involve GPU‑accelerated graph construction or incremental updates to maintain interactivity. Additional validation on other modalities such as ultrasound or PET, as well as integration into clinical workflow tools, are identified as next steps. Overall, Refinement‑Cut offers a practical bridge between fully manual delineation and fully automatic segmentation, delivering fast, accurate, and user‑controllable results for routine medical imaging tasks.