Open AccessSegmentation techniques for early cancer detection in red blood cells with deep learning‐based classifier—a comparative approach
Author(s) -
Shemona Jeya Sudharsan,
Chellappan Agees Kumar
Publication year - 2020
Publication title -
iet image processing
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.401
H-Index - 45
eISSN - 1751-9667
pISSN - 1751-9659
DOI - 10.1049/iet-ipr.2019.1067
Subject(s) - artificial intelligence , robustness (evolution) , segmentation , computer science , pattern recognition (psychology) , classifier (uml) , feature extraction , closeness , cancer detection , cancer , mathematics , medicine , biology , gene , mathematical analysis , biochemistry
Red Blood Corpuscles called Erythrocytes are the most important element in blood composition which is mainly responsible in all living cells. To detect the cancer cell various methods are employed. In this paper, proper identification of cancer cells from unaffected RBCs are detected. The proposed novel method called Online Region Based Segmentation (ORBS) method is done that is used to find the regions of corpuscles. By using properties, metric is formulated for determination of shape which is abnormal in blood cells. Overall accuracy of 96.9% is obtained using proposed ORBS method and deep learning classification (DLC) method has accuracy of 97.1% that helps to diagnose cancer cells using feature extraction process done automatically. Sensitivity, specificity and precision values of the proposed segmentation method is found to be 96.7%, 95.6% and 98.4% respectively. The computation time was found as 22 seconds. Closeness of Proposed method in relative to True Positive values at the ROC curve indicates the performance as higher. Comparative analysis is made with ResNet‐50 based on the different testing and training data at rate of 90%−10%, 80%−20% and 70%−30% respectively, which proves the robustness of proposed research work. Experimental results prove proposed system effectiveness compared with other detection methods.