Unmasking malignancy of lung nodule using a modernized ConvNet toward the design of a vision transformer

Abstract

One of the most devastating cancers in the world is lung cancer. It is estimated that nearly a third of the world’s cancer fatalities are due to lung cancer. Diagnosis and treatment of primary and metastatic cancers depend heavily on the ability to identify and characterize malignant cells. On the other hand, early detection of lung cancer is crucial for a patient’s survival and significantly increases the survival rate. Malignant lung nodules may be detected early by oncologists using a variety of diagnostic methods such as needle prick biopsy and other types of imaging tests such as CT and PET scanning, as well as clinical examinations and other types of imaging tests. It’s important to note that these treatments and biopsies are risky. A higher proportion of people are being infected with the disease, on the other hand. CT scans are commonly performed in the early stages of cancer detection. Lung cancer may be detected with a 2.6 to 10-fold higher CT detection rate than analog radiography, according to Awai [1]. As the slices get thinner, so does their ability to recognize objects accurately. To evaluate one slice, radiologists need an average of two to three minutes. The burden of cancer patients is increasing as the number of those diagnosed grows. CT imaging may be used to detect malignancy and cancerous nodules in a patient. When cancer nodules (stage I) are discovered, treatment may begin, and the danger of cancer spreading can be minimized. 70 percent to 92 percent of people diagnosed with stage 1 non-small cell lung cancer (NSCLC) should expect to live for at least five years following their diagnosis, according to existing statistics[30] . Considering the fact that a large number of early detection methods are already available, further research is needed to improve the accuracy of these methods and, as a consequence, the overall survival rate. Using ConvNeXt, we believe we can work more efficiently and precisely. Radiologists will also benefit greatly from this change. The validity of the proposed network was evaluated by comparing its performance to that of the other pre-trained CNNs, such as GoogleNet, AlexNet, and ResNet50, using a simulated dataset of pre-processed CT scan images: the Luna 16 dataset. Since our network outperforms the other networks in terms of classification, accuracy is evident from the results. Aside from pulmonary nodule detection, this proposal’s approach may be simply adjusted to conduct classification jobs on any 3D medical diagnostic computed tomography pictures where the classification is very unpredictable and ambiguous, such as any other 3D medical diagnostic CT images.