An Optimized CNN Framework for Automated Detection of Lung Cancer Using CT Scans

The dangerous illness that attacks human body respiratory system, lung cancer, and leaves a catastrophic effect on a person’s health and well-being. In the absence of automated and non-invasive diagnostic instruments, biopsy is anticipated by medical practitioners as the standard for diagnosis. However, the biopsy procedure can be costly and traumatic. Researchers also face significant challenges related to inaccurate diagnosis and restricted dataset availability. By utilizing optimum hyper-parameters, the research proposed aims in creating an automated diagnostic tool for lung cancer screening that would ensure the Convolutional Neural Network (CNN) model performs well for commonly collected computed tomography (CT) slices of lung diseases. The following methods are used to accomplish the above-specified goal. To prevent information loss from random picture smoothing, a pre-processing methodology tailored to lung CT scans is first developed. Secondly, a Sine Cosine Algorithm Optimization Algorithm (SCA) is incorporated into the CNN model to help choose the CNN tuning parameters in the best possible way. The SCA algorithm uses the error rate as an objective function that it seeks to minimize. The suggested approach effectively classified lung scans into groups such as normal, benign, and malignant with an average classification accuracy of 99%, demonstrating the system’s suitability for use by radiologists in a clinical setting.