Design and compare CNN deep learning model to classify lung cancer using CT images

Abstract

Cellular breakdown in the lungs has become one of the most widely recognized passings
among the disease patients. World Health Organization expresses that cellular breakdown in
the lungs is the second most lethal malignant growth all around the world in 2014.
Alarmingly, a large portion of the cellular breakdown in the lungs patients are analyzed at the
later stages where the disease has spreads. Hence, early screening by means of Computed
Tomography check especially among dynamic smokers is energized. Manual determination
of the malignant growth is made practical through the coordination of Computer Aided
Diagnosis framework. For the beyond couple of years, profound learning strategy drives the
greater part of the fake based knowledge applications including CAD frameworks. This paper
means to research the presentation of five recently settled Convolutional Neural Network
designs; GoogleNet, SqueezeNet, DenseNet, ShuffleNet and MobileNetV2 to order lung
cancers into dangerous and harmless classes utilizing LIDC-IDRI datasets. Their exhibitions
are estimated as far as exactness, responsiveness, explicitness and region under the bend of
the recipient working trademark bend. Trial results show that GoogleNet is the best CNN
design for CT lung cancer arrangement wih an exactness of 94.53%, explicitness 99.06%,
responsiveness of 65.67% and AUC 86.84%.