Lung cancer detection using enhanced segmentation accuracy
- PDF / 2,295,781 Bytes
- 14 Pages / 595.276 x 790.866 pts Page_size
- 65 Downloads / 188 Views
Lung cancer detection using enhanced segmentation accuracy Onika Akter 1 & Mohammad Ali Moni 2,3,4 & Mohammad Mahfuzul Islam 5 & Julian M. W. Quinn 4,6 & A. H. M. Kamal 1 Accepted: 28 October 2020 # Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Lung cancer is currently one of the most common causes of cancer-related death. Detecting and providing an accurate diagnosis of potentially cancerous lung nodules at an early stage of their development would increase treatment efficacy and so reduce lung cancer mortality. A key barrier to early detection is the absence of noticeable symptoms until the lung cancer has already spread. Diagnosis and screening using non-invasive imaging such as computed tomography (CT) is a potential solution. However, to realize the potential of this approach an accurate automated analysis of these high-resolution images needed. Image segmentation is an important stage of that process. Fuzzy-based image segmentation schemes use the maximum of each row and minimum of each column. Our study developed an algorithm that employs median values measured along each row and column, in addition to the maxima and minima values, and found that this approach increased segmenting accuracy of these images,. In the next phase of analysis, a neuro-fuzzy classifier classified those segmented lung nodules into malignant and benign nodules. Sensitivity, specificity and accuracy were used as performance assessment parameters. The proposed methodology resulted in sensitivity, specificity, precision and accuracy of 100%, 81%, 86% and 90%, respectively, with a reduced false positive rate. In sum, our improved algorithm can give significantly improved accuracy of diagnosis in early-stage patients from CT imaging. Thus, our methodology could contribute to better clinical outcomes for lung cancer patients. Keywords Lung cancer . Computer tomography . Morphology . Image segmentation . Nodule feature extraction . Cluster . Classification
1 Introduction Lung cancers, including small cell and non-small cell carcinomas, are relatively common in most parts of the world and have a very high mortality rate. They present with uncontrolled growth of abnormal neoplastic cells in the lungs that not only compromise lung function but rapidly metastasise to many other sites, and this can have fatal consequences if not treated at the earliest possible stage [1]. However, most types of lung cancers can be very difficult to detect at an earlier stage
of progression because symptoms usually become evident only when the condition has reached an advanced stage of invasive and metastatic growth. Accurate early diagnosis can also be difficult due to the soft nature of pulmonary tissue which makes accurate biopsies and X-ray-based imaging problematic. In addition, while a bronchoscopy and lung biopsy can identify tumour cells correctly (Fig. 1) both are painful invasive procedures that are not undertaken lightly. Without convincing evidence of lung cancer, physicians are unlikely to suggest performing a biopsy
Data Loading...
