Loading [a11y]/accessibility-menu.js
Image classification of lung nodules by requiring the integration of Attention Mechanism into ResNet model | IEEE Conference Publication | IEEE Xplore

Image classification of lung nodules by requiring the integration of Attention Mechanism into ResNet model


Abstract:

In this research, in order to accurately diagnose lung nodules using the LUNA16 dataset, a deep learning model, ResNetl01, is analyzed and chosen. The paper includes: (1)...Show More

Abstract:

In this research, in order to accurately diagnose lung nodules using the LUNA16 dataset, a deep learning model, ResNetl01, is analyzed and chosen. The paper includes: (1) demonstrating the efficiency of the ResNetl01 network on the LUNA16; (2) analyzing the benefits and drawbacks of Attention modules before selecting the best Attention module to integrate into the ResNetl01 model in the classification of lung nodules in CT scans challenge; (3) comparing the efficacy of the proposed model to prior outcomes to demonstrate the model’s feasibility.
Date of Conference: 19-21 October 2022
Date Added to IEEE Xplore: 21 November 2022
ISBN Information:

ISSN Information:

Conference Location: Nha Trang, Vietnam
Related Articles are not available for this document.

I. Introduction

Lung cancer is not new to us, with silent symptoms in the early stages but outbreaks in the late-stage. According to statistics, about 80% of lung cancer patients are diagnosed when the disease is at a late stage with clinical symptoms such as cough, chest pain, shortness of breath, weight loss, etc. In 2017, lung cancer killed more people than breast, prostate, colorectal, and brain cancers combined [1]. The detection of small tumors may offer hope to patients with a treatment that combines surgery and chemotherapy. Routine screening with general tests or noninvasive imaging modalities like low-dose computed tomography (CT) aids in the early detection of lung nodules when they are small and show no evidence of metastasis. However, imaging diagnosis puts pressure on inexperienced doctors. Computer aided system with automatic tumor identification algorithms is really needed. These algorithms are categorized into two groups: those that employ handdraft features and those that use deep learning networks. For example, one approach is to use the Gabor algorithm to segment the image area as a fingerprint reference point or not [2]. These techniques can also be applied to clarify the tumor or non-tumor areas in medical images, however, due to the diverse and complex nature of medical images, using Gabor will speed up the treatment. image processing is slower, so we decided to use the most modern trends in the deep learning family to solve this problem. On the other hand, there are many groundbreaking studies applying deep learning networks to solve the problem of identifying tumors in the lung. Dou and colleagues [3] create three deep learning network designs (Archi-1, Archi-2, and Archi-3) on the base of the multi-context 3D convolutional network idea known as CUMedVis to overcome issues caused by variances in node size, shape, and geometrical data. These three designs are then combined with a weighted linear combination to provide the final classification result for the candidates. Ding [4] suggested a CNN-based pulmonary node identification method. Their technique use region-based CNNs to detect nodules on image slices and 3-D CNNs to decrease false positives. The approach was tested on the LUNA16 dataset and shown great sensitivity (94.4 %).

References

References is not available for this document.