Innovate histopathological image diagnosis using deep machine learning image analysis

Renal cancer is one of the top causes of cancer-related deaths worldwide, and early detection of renal cancer can significantly improve the patients’ survival rate. However, the manual analysis of renal tissue in the current clinical practices is labour-intensive and prone to inter-pathologist varia...

全面介绍

Saved in:
书目详细资料
主要作者: Koo, Jia Chun
格式: Final Year Project / Dissertation / Thesis
出版: 2022
主题:
在线阅读:http://eprints.utar.edu.my/5236/1/BI_1702882_Final_%2D_JIA_CHUN_KOO.pdf
http://eprints.utar.edu.my/5236/
标签: 添加标签
没有标签, 成为第一个标记此记录!
实物特征
总结:Renal cancer is one of the top causes of cancer-related deaths worldwide, and early detection of renal cancer can significantly improve the patients’ survival rate. However, the manual analysis of renal tissue in the current clinical practices is labour-intensive and prone to inter-pathologist variations. In this work, we have developed deep convolutional neural network (CNN) based heterogeneous ensemble models for automated analysis of renal histopathological images. The proposed method in the first step segmentate the histopathological tissue into patches with different magnification factors, whereas in the second step, classify the generated patches into normal and tumour tissues using the pre-trained CNN and ensembled models. For the development of heterogeneous ensemble models, CNN models from five deep learning architectures, namely VGG, ResNet, DenseNet, MobileNet, and EfficientNet, are fine-tuned and used as base learners. These CNN models exhibit different performances and have great diversity in histopathological image analysis. The CNN models with superior classification accuracy are selected to undergo ensemble learning for further improvement. The performance of the investigated ensemble approaches is evaluated against the state-of-the-art literature. The performance evaluation demonstrated the superiority of the best proposed ensembled model: fiveCNN based weighted averaging model, with higher accuracy (99%), specificity (98%), F1-score (99%) and area under the receiver operating characteristic curve (98%) but slightly inferior recall (99%) compared to the literature. The outstanding robustness of the developed ensemble model with high performance scores in the evaluated metrics suggests its reliability as a diagnostic system for assisting the pathologists in analysing the renal histopathological tissues. It is expected that the proposed ensemble deep CNN models can greatly improve the early detection of renal cancer and subsequently, leading to higher patients’ survival rate