A study of Accuracy in Detection of Lung Cancer through CNN Models

  • Nikhelesh Bhattacharyya, Ajay Sudharshan Satish

Abstract

Purpose –The purpose of this paper is to study the five universally recognised convoluted neural network models (VGG16, VGG19, ResNet50, Inception V3) and by using concepts of data mining, image processing and deep learning to generate predictions, compare their results, discuss their benefits and limitations and decide which model is the most accurate in prediction of Lung Cancer from Histopathological images.

Design/methodology/approach – The paper is primarily based on an image dataset ‘Borkowski AA, Bui MM, Thomas LB, Wilson CP, DeLand LA, Mastorides SM. Lung and Colon Cancer Histopathological Image Dataset (LC25000). arXiv:1912.12142v1 [eess.IV], 2019’ taken from Kaggle. The data contains Histopathological images of the lungs. The images were first preprocessed and then added to the Model. The CNN models were created using Keras Library along with concepts the of transfer learning and fine turing.

Findings – Out of the five models used, VGG16, VGG19 and ResNet 50 have a good accuracy with ResNet 50 being the highest, narrowly followed by VGG16 and VGG19, while Inception V3 gives the least accuracy and is unfit for prediction of Lung Cancer from Histopathological images.

Practical implications – This study aims to study the use of convolutional neural networks to effectively diagose cancer and how convolutional neural networks models can continously evolve and improve to return highly accurate predictions.

Originality/value – There are several studies on the use of convolution neural network  models and their application in medical fields. This study aims to compare the five most recognized CNN models, modify them by using concepts of transfer learning and fine-tuning, and understand which model can accurately diagose Lung cancer from Histopathological images

Published
2021-12-01
How to Cite
Nikhelesh Bhattacharyya, Ajay Sudharshan Satish. (2021). A study of Accuracy in Detection of Lung Cancer through CNN Models. Design Engineering, 1250 - 1270. Retrieved from http://www.thedesignengineering.com/index.php/DE/article/view/7073
Section
Articles