Advanced imaging techniques for lung cancer diagnosis with deep learning

Abstract

Cancer is a major public health problem worldwide that kills millions of people globally with lung cancer among the commonest noncommunicable diseases in Bangladesh, causing 6% of all deaths It can present with a wide varying symptoms ranging from chest and bone pain, wheezing, hoarseness, persistent chest infections, unexplained weight loss, fatigue, changes in the appearance of fingers known as finger clubbing, shortness of breath and recurrent pneumonia to hemoptysis. Several risk factors are associated with lung cancer and include, long term smoking, exposure to secondhand smoke, exposure to radon gas, exposure to asbestos and other carcinogens, radiation therapy to the breast or chest, family history of lung cancer and so on. However, the constraint of the traditional CT is that it is not ideal for early detection, unable to meet “the requirements of population level screening because of the associated high false positive rates,” the researchers explained and an uncomfortable procedure for the patient. In fact, the approach can enhance the earlystage detection, potentially raising survival rates while also reducing the number of unnecessary follow-up tests and treatments that could otherwise add up over time, where the return on investment is quite substantial. In the presented work, we aimed to identify only lung cancer in CT scan data of the lung and so our project is focused on lung cancer only. Using binary 0 and 1, we have developed a system which detects if a picture is malignant or benign and it could be used in hospitals where patients with lung cancer are being treated/seek treatment. The experimental results indicated that all the models are not achieved high accuracy, with most models surpassing 96% accuracy. Notably, the CNN and DenseNet121 models performed exceptionally well, demonstrating robust capabilities in accurately detecting lung cancer from CT scan images.