الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 136 |  |  |
| | | from | 136 | | |
Abstract
Lung and colon cancer have the most severe death and incidence rates of all the common malignancies across the world. Early diagnosis of the disease increases survival chances for affected people. An important element of cancer type identification is histopathological diagnosis. It is urgently necessary to analyze histopathological images of lung and colon cancers since the type of histology, molecular profile, and stage of diagnosis all influence how the disease is treated. Pathologists can use deep learning methods to diagnose lung and colon cancer more quickly and reduce stress. It aims to make computers capable of analyzing, identifying, and perceiving images similarly to humans. Additionally, it produces the desired results. It is like giving a machine-human intelligence and instinct. Convolutional neural networks (CNN), in particular, enhance efficiency in cancer histology slide examination.
The thesis utilizes a CNN model to extract features from images after undergoing preprocessing, followed by the classification of lung and colon tissues using an improved Light Gradient Boosting Machine (LightGBM). The effectiveness of the proposed CNN feature extraction model is compared to other deep learning methods, such as VGG16, VGG19, AlexNet, Inception ResNet v2, ResNet50, Inception v3, GoogLeNet, and MobileNet. Additionally, the performance of the LightGBM classifier is evaluated against other machine learning models, including KNN, SVM, RF, AbaBoost, and XGBoost. Test images are separated into benign and malignant lesions without human intervention.
Furthermore, the suggested CNN model boasts the lowest training parameters with just one million parameters. The proposed CNN-LightGBM model outperforms other state-of-the-art approaches in the feature extraction and classification of lung and colon cancer histopathological datasets. The suggested approach achieves feature extraction and classification within three seconds. This points towards a superior diagnostic speed and high accuracy in disease classification compared to more recent methods.