1：Experimental Design and Method Selection:

The method extends a general simultaneous segmentation and registration framework based on Markov Random Field (MRF), using a multi-resolution scheme with graph cuts optimization via α-expansion. It involves a two-layer graph for segmentation and registration connected by coherence edges, with B-spline interpolation for deformation field prior.

2：Sample Selection and Data Sources:

Experiments were performed on the DIR-Lab data set, which contains 10 pairs of 4D lung CT scans with complete breathing cycles of patients suffering from lung or esophageal cancer. The first five cases were used, with slice thickness of 2.5mm and in-plane resolution from 0.97mm to 1.16mm. Landmarks annotated by medical experts were used for evaluation.

3：5mm and in-plane resolution from 97mm to 16mm. Landmarks annotated by medical experts were used for evaluation. List of Experimental Equipment and Materials:

3. List of Experimental Equipment and Materials: Not specified in the paper.

4：Experimental Procedures and Operational Workflow:

The process starts from a coarse control grid (24mm), uses multi-level framework (4 levels), and involves calculating MRF energy, optimizing with graph cuts, interpolating deformation with B-spline, and iterating for precise displacements. Parameters were set empirically: θreg=1, λreg=0.01, θseg=0.1, θcoh=

5：01, θseg=1, θcoh=Data Analysis Methods:

1. 5. Data Analysis Methods: Registration accuracy was measured by Target Registration Error (TRE) using Euclidean distance between landmarks. Segmentation accuracy was evaluated using DICE coefficient. Comparisons were made with methods like MRF-Reg, GCV+Vem, Vemuri, and MRF-Seg.