Automatic patient-customised 3D reconstruction of human costal cartilage from lung MDCT dataset.

Abstract

The costal cartilage is a prominent feature of the anterior chest wall that is subject to developmental and acquired abnormalities. A fully automatic algorithm to reconstruct the human costal cartilage from multidetector computed tomography (MDCT) images was developed and tested. The reconstruction algorithm includes three steps: (1) estimation of length, curvature and end points for each costal cartilage centre-line, (2) costal cartilage cross-section area approximation, and (3) transformation of the estimated cross-section to the centre-line into a cylindrical coordinate system. Four different models were as follows: circle, vertical ellipse, horizontal ellipse and a non-geometric shape have been assumed for the cross-section. Shape estimates were based on each patient's dataset, so the algorithm is patient-specific and anatomically faithful. MDCT datasets from 15 patients were evaluated with the automated algorithm and the results compared with reference masks provided by an experienced radiologist. The costal cartilage reconstruction result and the reference mask were visually consistent. Based on evaluation results, the circular model cross-section with area of twice M (mean area of all rib cross-sections in the mid-coronal plane) had the highest Dice similarity coefficient (DSC = 77.5%) with only 2.12 mm registration distance. Costal cartilage 3D morphology can be extracted from MDCT scans with an automated method, using a circular cross-section with area equal to twice M.