Background:
The most common histopathologic malignant and benign nodules are Adenocarcinoma and Granuloma, respectively, which have different standards of care. However, distinguishing Adenocarcinoma from Granuloma in the CT scan of the chest is a challenging task, due to the similar appearance in shape and appearance. Indeed, biopsies are needed for the diagnosis. The radiomic features of pulmonary nodules, along with the torsion of the vessels attached to the nodules are accepted by expert radiologists as the biomarker for discriminating the benign nodules from the malignant ones. In this paper, we propose an automatic framework for the distinction of the Adenocarcinomas and the Granulomas in CTs using the radiomic features of nodules and the attached vessel tortuosity.
Methods:
To automatically segment the delineated nodule area and the attached vessels area, we apply a morphological-based approach. For distinguishing the malignancy of the segmented nodule, two texture features of the nodule, the curvature mean and the number of the attached vessels are extracted. Then, we apply a trained SVM classifier to identify the segmented nodule as the Adenocarcinoma or the Granuloma.
Results:
The proposed framework is evaluated on a private dataset, including 22 CTs for each nodule type, i.e., adenocarcinoma and granuloma (44 CTs for both nodule types). The dataset contains the CTs of the non-smoker patients who are between 30 and 60 years old. Compared to the state-of-the-art feature selection methods which employ only the shape features of the nodule, the texture features of the nodule, or the torsion features of the attached vessels along with the radiomic features of the nodule for differentiating Adenocarcinomas from Granulomas, the accuracy of our framework is improved by 21:39%, 3:72%, and 8:27%, respectively. The area under the ROC curve of the introduced framework for the manually and automatically segmented nodules is 0:8874 and 0:7583, respectively.
Conclusions:
The AUC value for the automatically segmented nodules is lower than that of the manual ones labeled by a radiologist. However, the time run of the introduced framework for the automatically segmented nodules is much lower than that of the manual ones.