Development of 3DTDUnet++ with novel function and multi-scale dilated-based deep learning model for lung cancer diagnosis using CT images

Abstract

The early diagnosis of lung cancer performs a vital role in enabling doctors to provide timely treatment, potentially saving patients' lives. However, the intricate nature of cancer in lung features makes computer-aided automatic diagnosis a particularly challenging task. Unfortunately, the implications of deep structures in the realm of radiology diagnosis remain somewhat constrained, principally due to the shortage of extensive medical image data domains. Nevertheless, computer-aided diagnosis systems hold immense promise in assisting medical professionals in accurately identifying cancerous cells. Over time, various computer-aided techniques that are involved in image processing and machine learning have been extensively researched and implemented with the aim of enhancing diagnostic accuracy. Hence, this paper aims to develop an automated diagnosis model for cancer in lungs using CT images. The foremost step is the collection of CT images from the respective data domains. Further, the images are subjected to the model of 3D Trans-DenseUnet++ with Novel Loss Function (3D-TD ++-NLF) for segmenting the abnormal regions. At last, with the help of segmented images, lung cancer is diagnosed by proposing the Multi-scale Dilated 3D DenseNet with Atrous Spatial Pyramid Pooling (MDDNet-ASPP). The suggested model is examined and evaluated with distinct performance measures. From the result analysis, the accuracy and precision rate of the designed approach are 92.925 and 93. Therefore, extensive results are obtained to prove the prior diagnosis of cancer in the lungs and also aid the practitioner in treating the patient at a very early stage.