Currently, evaluation of abnormal lesions on lung Computed
Tomography (CT) images is an important step, especially in
patients who have tumor in the early stages, leading to increased survival rates. In early cases of tumor diagnosis on lung, Positron Emission Tomography (PET-CT) and histology images (colored) are very complicated since the intensity values of healthy and abnormal tissues may be very close. The objective of this paper is to differentiate between healthy and abnormal tissues through an image processing clustering algorithm. Fuzzy c-means clustering algorithm is applied to the lung PET-CT and histology images. The algorithm uses the microscopic examination of malignant and benign tissues to improve clustering process based on minimization of the objective function. This paper introduces a new method for predicting the type of patients with unknown lung cancer from their PET-CT images in early stages. The proposed technique differentiates between normal and abnormal tissues based on histopathological information. This paper develops a membership function based on iterative optimization to find the similarity between any measured data and the center leading to improving the clustering process. This incorporates preprocessing stages of noise removal and image enhancement.
The diagnosis stage includes color PET-CT and histology image segmentation to identify the region with abnormal tissue. This leads to improved early diagnosis of lung cancer. Finally, the proposed technique measures the percentage of affected area with cancerous tissue. The algorithm is applied to 40 sets of different real data in the form of lung PET-CT and histology
images with normal, abnormal tissue and early tumor. The
experimental results show that the proposed algorithm proved effective in detecting tumors on lung PET-CT especially in
images having tumors that were undetected by traditional
methods. |
