Abstract
This paper presents a scheme to analyze surrounding structures of pulmonary nodules by using differential geometry based vector fields. In this scheme the differential characteristics such as the principal curvatures and directions are computed from the differential values of the isointensity surfaces. Each voxel in the nodule surrounding is described in terms of shape index and curvedness derived from the principal curvatures. Two vector fields are formed from the directions of the maximum principal curvatures of nodule surrounding and gradient vectors of nodule surface, respectively. The gradient vector field is computed by diffusing the gradient vector on the nodule surface. The regions corresponding to the cylindrical or conic figures which are similar to vessel and plural images are segmented by the shape index and curvedness values. Then, the relationship between the segmented regions and the nodule is evaluated by the inner product of the direction of the maximum principal curvature and the gradient vector. We demonstrate the feasibility of the scheme to classify benign and malignant nodules.
Introduction
Distinguishing between benign and malignant pulmonary nodules is a high-stakes challenge in radiology. While thin-section CT scans provide a non-invasive look at these small growths, visual inspection alone isn't always enough to make a definitive diagnosis. Clinicians have long noted that malignant tumors often exhibit "vascular convergence" (vessels appearing to point toward the tumor) and "pleural retraction" (the tumor pulling on the lung lining). However, turning these qualitative observations into a reliable, automated computer-aided diagnosis (CAD) tool is difficult. This paper addresses the need for a mathematical framework that can precisely measure these structural relationships in 3D space by looking beyond the nodule itself to the interaction with its environment.
Methods
The proposed scheme utilizes differential geometry to quantify the relationship between a nodule and its surrounding structures. The process involves several sophisticated steps:
- Segmentation and Curvature Analysis: The nodule and its immediate surroundings are isolated from 3D CT data. For every voxel in the surrounding area, the shape index (identifying structures like tubes or ridges) and curvedness (measuring sharpness) are calculated. This helps identify tube-like structures such as blood vessels.
- Vector Field Construction: Two distinct vector fields are created. The Maximum Principal Curvature Vector (MPV) field identifies the axial direction of tube-like structures. Simultaneously, a Gradient Vector (GV) field is generated by diffusing the gradient of the nodule's surface outward, representing the direction pointing directly toward or away from the nodule.
- Convergence Metric ($F$): The alignment between these fields is evaluated using the inner product of the MPV and GV fields. A high $F$ value indicates that a surrounding vessel is pointing directly toward the nodule, a key indicator of malignancy.
Datasets
The study utilized a clinically verified dataset of 20 3-D thoracic CT images from the National Cancer Center Hospital East. This included 10 malignant cases (mostly adenocarcinomas confirmed via cytology or histology) and 10 benign cases (including inflammatory nodules and tumor-like lesions). Images were reconstructed to ensure isotropic voxels for accurate 3D analysis.
Results
The researchers introduced a feature called $F1$, representing the ratio of voxels in the surrounding area showing high convergence toward the nodule. Malignant nodules exhibited significantly higher $F1$ values than benign ones ($p < 0.01$). The system was particularly effective at distinguishing malignant tumors from "tumor-like" benign lesions. Histograms of the $F$ values clearly showed a "heavy tail" in malignant cases, indicating a much higher frequency of vessels pointing directly at the mass compared to benign cases.
Discussion and Future Work
While the results are promising, the study serves as a proof of concept due to its small sample size. The system occasionally struggled with inflammatory benign nodules that mimic the "pulling" effect of malignancies. Future work will focus on automated parameter adjustment for Gaussian filters (currently set at $\sigma = 2.0$) and integrating these vector field features into a larger feature space that includes internal nodule density and texture to improve diagnostic accuracy.
Summary updated