Efficient Lung Cancer Classification on Multi level Convolution Neural Network using Histopathological Images
M. Ramesh, S Maheswaran, S Theivanayaki, K Kodeeswari, Lalitha Krishnasamy, N S Sriram
Abstract
Lung cancer can be detected by lung nodules, which are a key sign. An early diagnosis enhances the likelihood that the patient will survive by enabling the appropriate therapy to start. To reduce the responsibility of radiologists' difficult and time-consuming labour of finding and categorising malignancy in Computed Tomography (CT) images, researchers have created CAD (computer-assisted diagnosis) systems. The likelihood and kind of malignancy are commonly determined by pathologists using histopathological images of biopsy specimens taken from potentially sick areas of the lungs. To categorise lung nodule malignancy, we recommend employing a four-level convolutional neural network (ML-CNN). From lung nodule CT scan images, multiple scales are extracted. ML-CNN’s employs four CNNs network model structure. After the result of the last pooling layer has been flattened to a vector with a single dimension for each level, the vectors are concatenated. These four ML-CNNs will help our model perform better. The ML-CNN model can recognise and classify different forms of lung cancer with reasonable accuracy. The 25000 images employed in the ML-CNN model have been separated into three categories: training, validation, and testing. Three distinct tissue types were assessed and training and validation took up within 80% and 15% of the total time and 5% for testing, respectively. The histopathological images included the following tissue type’s 1.Benign tissue 2. Large cell carcinoma 3.squamous cell carcinoma. The proposed model demonstrated superior performance on both the training set, achieving an accuracy of 78%, and the validation set, achieving an accuracy of 89.6% by the end of the evaluation