Affiliation:
1. School of Computer Science and Engineering, Vellore Institute of Technology, Vellore, India
Abstract
Identifying diseases using chest X-rays is challenging because more medical professionals are needed. A chest X-ray contains many features, making it difficult to pinpoint the factors causing a disease. Moreover, healthy individuals are more common than those with illnesses, and various diseases occur at different rates. To diagnose the disease accurately using X-ray images, extracting significant features and addressing unbalanced data is essential. To resolve these challenges, a proposed ensemble self-attention-based deep neural network aims to tackle the problem of unbalanced information distribution by creating a new goal factor. Additionally, the InceptionV3 architecture is trained to identify significant features. The proposed objective function is a performance metric that adjusts the ratio of positive to negative instances, and the suggested loss function can dynamically mitigate the impact of many negative observations by reducing each cross-entropy term by a variable amount. Tests have shown that ensemble self-attention performs well on the ChestXray14 dataset, especially regarding the dimension around the recipient’s characteristics curves.
Reference34 articles.
1. Huang G. , Liu Z. , Weinberger K.Q. , van der Maaten L. , Densely connected convolutional networks Available: https://arxiv.org/abs/1608.06993.
2. Rethinking the inception architecture for computer vision;Szegedy;Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR),2016
3. ChestX-ray8: Hospital-scale chest X-ray database and benchmarks on weakly-supervised classification and localization of common thorax disease;Wang;Proc. CVPR,2017
4. Imagenet classification with deep convolutional neural networks;Krizhevsky;Communications of the ACM,2017
5. Esteva A. , et al., Dermatologist-level classification of skin cancer with deep neural networks, Nature 542(7639) (2017), 115–118. Available: https://www.nature.com/articles/nature21056.