A Modified Dense-UNet for Pulmonary Nodule Segmentation

• Autor(a) principal: Naqvi, Najme Zehra
• Data de Publicação: 2024
• Outros Autores: Chhikara, Muskaan, Garg, Arushi, Yashika, Agrawal, Milan
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: INFOCOMP: Jornal de Ciência da Computação
• Texto Completo: https://infocomp.dcc.ufla.br/index.php/infocomp/article/view/2886

Lung cancer continues to be a major health concern worldwide, taking countless lives every year. Although the detection of lung nodules has been made versatile by using CT scans, radiologists require certain assistance to make this process faster and more efficient. This need led to the introduction of Computer Aided Diagnosis (CAD) and then, deep learning into the healthcare field. In this work, we have proposed a modified 2D Dense-UNet model for the segmentation of lung nodules from the CT scan images. The model is trained and tested on the LUNA16 dataset which is publicly available. Through the addition of Squeeze & Excitation (SE) blocks and the GeLU activation function in its dense layers, some improvement has been observed in the basic model. Furthermore, we have also compared our suggested model's performance to that of various other 2D deep learning networks on the basis of their Dice Coefficient (DSC).