Lung cancer is one of the major causes of non-accidental death. Because the lung is in touch with the external environment, it is affected most of all human viscera. According to a recent WHO report, due to the environmental pollution by overurbanization and industrialization, lung cancer is the leading cause of cancer-related deaths worldwide.
Now, the common and basis tool for early diagnosis of lung cancer is a chest CT machine. Modern CT imaging has made much progress in terms of resolution, speed and clarity compared to previous CT imaging with the advent of technology including information technology, but it still reveals some limitations in the perfect diagnosis of lung-related diseases from chest CT images. The diagnosis of lung cancer in the early stage requires biopsy and analysis of CT images taken at intervals of 6 to 18 months, which is dangerous and uncertain. However, many researchers have reported that it is possible to detect early stage lung cancer from CT images for timely treatment.
In particular, many studies have been carried out to develop a computer-aided detection/diagnosis system (CADe/CAD) that can help physicians with diagnosis by automatically detecting areas of suspected pulmonary nodules from the resulting large number of CT images, but there are still challenges to directly apply CADe/CAD for early detection of lung cancer due to the complexity of the lung and the resolution of CT images.
Based on the growth characteristics of cancer tissue cells, Pak Chun Hyok, a section head at the Faculty of Electronics, has proposed a method of detecting malignant pulmonary nodules with high ambiguity and high probability of developing into lung cancer.
The proposed scheme consists of four major stages: extraction of lung region from CT images, extraction of the region of interest (ROI), detection of candidate early lung cancer regions among extracted ROIs, and detection of early lung cancer regions.
The details are as follows.
First, lung segmentation using superpixels is performed to detect the correct lung region.
Second, based on the merging between superpixels within the obtained lung region, ROI is obtained.
Third, image segmentation based on level set algorithm and superpixel segmentation is performed to evaluate the ambiguity of candidate lesion regions.
Finally, malignant nodule is extracted by 3D continuity of lung cancer.
The evaluation result shows that the proposed method is feasible and effective for detection of malignant nodules.
© 2021 Kim Chaek University of Technology