A study of lung cancer detection using computer aided detection algorithms for various imaging modalities

Abstract

Lung cancer is considered as the notable cancer because it claims more than a million lives every year. The requirement of techniques to identify the occurrence of this cancer disease in the beginning phase is very much essential. In general, Computer Aided Detection (CAD) is computer software which are used to highlight suspicious features on an image and bring them to the attention of the radiologist. Computer Aided Detection (CAD) algorithm plays a vital role in the study of lung cancer detection for the issues faced with an existing system using three images which are listed in parentheses (X-Ray, Computed Tomography (CT), Positron Emission Tomography (PET)). These images are downloaded from appropriate dataset (Japanese Society of Radiological Technology, Lung Image Database Consortium/Image Database Resource Initiative, Anderson Diagnostics and Labs). An issue faced by an existing system using X-Ray images is an overlap of rib and clavicles with lung nodules. It fails to detect the subtle nodules due to increase in false positive which leads lower sensitivity and accuracy. A challenging issue faced by the existing system using CT and PET images is an increase in false positive due to variation in the pulmonary nodule. It arises due to low contrast in Computed Tomography image and low spatial resolution in Positron Emission Tomography image. The list of CAD algorithm used for the study of lung cancer detection in beginning phase is Massive Artificial Neural Network (MANN) based soft tissue technique, Ten Convolutional Neural Network (CNN) models with three classifiers and Modified co-learning technique based on ten CNN models. The MANN is a non linear filter used to accommodate the task of distinguishing a specific opacity from other opacities and the CNN is a type of artificial neural network used primarily for image recognition and processing with computational ability. In an X-Ray image, the subtle nodules of various types are detected using MANN filter. By using these images, CAD Algorithm attained 72.85% sensitivity and 72.96% accuracy by using the parameters (True Positive, True Negative, False Positive and False Negative) available in a confusion matrix