A novel attention fusion network-based framework to ensemble the predictions of CNNs for lymph node metastasis detection
- PDF / 1,739,725 Bytes
- 20 Pages / 439.37 x 666.142 pts Page_size
- 60 Downloads / 161 Views
A novel attention fusion network‑based framework to ensemble the predictions of CNNs for lymph node metastasis detection Chinmay Rane1 · Raj Mehrotra1 · Shubham Bhattacharyya1 · Mukta Sharma1 · Mahua Bhattacharya1 Accepted: 4 September 2020 © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Diagnosis of lymph node metastases is a challenging task for pathologists, involving an extensive screening of the pathological scans. Automating diagnostic processes reduces the workload of pathologists and yields high accuracy by the virtue of advances in technology. In this study, a novel ensemble-based framework is proposed for the classification of lymph node metastases. The proposed ensemble framework comprises different pre-trained CNN models such as DenseNet201, InceptionV3 and ResNeXt-50. In the proposed framework, an attention fusion network is utilized to amalgamate the predictions of the individual models. The proposed framework achieves an AUC-ROC of 0.9816 which surpasses the highest AUC-ROC achieved by the conventional approaches on the PCam benchmark dataset. Keywords Attention fusion network · Cancer detection · Convolutional neural network · Ensemble · Progressive resizing · Test time augmentation
* Mukta Sharma mukta.24sharma@gmail.com Chinmay Rane chinmayrane16@gmail.com Raj Mehrotra mehrotraraj03@gmail.com Shubham Bhattacharyya tutushubham@gmail.com Mahua Bhattacharya mahuabhatta@gmail.com 1
ABV-Indian Institute of Information Technology and Management, Gwalior, MP, India
13
Vol.:(0123456789)
C. Rane et al.
1 Introduction Cancer is recognized as the second leading cause of deaths worldwide according to the World Health Organization [1]. In 2018, cancer claimed over 9.6 million deaths globally. Cancer is a disease with a characterizing highlight of the quick formation of anomalous cells that develop past their typical limits. It is also sometimes referred to as malignant tumors or neoplasms. These cancer cells leave the original site of the tumor called the primary tumor. Furthermore, it flows through either the circulation system or lymphatic system and forms a malignant growth at some other site. This is called metastasis [2]. The type of metastatic tumor is the same as the primary tumor and follows similar treatment procedures [3]. Classification of metastases is an arduous exercise even for skilled pathologists because it involves identifying even a single cell affected with a tumor under the microscope. In addition, the heterogeneity and contradiction among pathologists diagnosis further convolute the task. The assessments of the pathologists vary nearly 20% of the time [4]. The variations in textures, structures and heterogeneous appearance cause the manual diagnosis of histopathological images to be more challenging [5]. The likelihood of diagnostic error is elevated due to the inherent tedious and fatigue nature of the task. Owing to these issues, there exists an increasing demand for skilled pathologists. Automation of the detection of metastases becomes nec
