TY - GEN
T1 - 1D-DCGAN for Oversampling Minority Mitotic Patterns in HEp-2 Cell Images
AU - Anaam, Asaad
AU - Al-Masni, Mohammed A.
AU - Gofuku, Akio
N1 - Publisher Copyright:
© 2022 IEEE.
PY - 2022
Y1 - 2022
N2 - This paper proposes a framework for oversampling the minority mitotic patterns of the HEp-2 cell images. The classification of mitotic vs. non-mitotic (interphase) cell patterns is important for validating the Indirect Immunofluorescence on Human Epithelial Type-2 cell-substrate (IIF HEp-2) protocol, which is the "gold standard"test for diagnosing autoimmune diseases. Typically, the mitotic cells appear in the HEp-2 specimen image in a significantly less number than the interphases. This causes difficulty in adopting deep learning approaches to classify mitotic vs. interphase patterns with such high imbalanced data. This work suggests using One-Dimensional Deep Convolutional Generative Adversarial Networks (1D-DCGAN) for oversampling the minority mitotic patterns in the feature space of the Deep Cross Residual Network (DCRNet) to cope with the data skewness problem. The results demonstrated that the proposed approach improves the classification performance over the UQ-SNP I3A Task-3 mitotic cell dataset with the advantage of using an end-to-end CNN classifier.
AB - This paper proposes a framework for oversampling the minority mitotic patterns of the HEp-2 cell images. The classification of mitotic vs. non-mitotic (interphase) cell patterns is important for validating the Indirect Immunofluorescence on Human Epithelial Type-2 cell-substrate (IIF HEp-2) protocol, which is the "gold standard"test for diagnosing autoimmune diseases. Typically, the mitotic cells appear in the HEp-2 specimen image in a significantly less number than the interphases. This causes difficulty in adopting deep learning approaches to classify mitotic vs. interphase patterns with such high imbalanced data. This work suggests using One-Dimensional Deep Convolutional Generative Adversarial Networks (1D-DCGAN) for oversampling the minority mitotic patterns in the feature space of the Deep Cross Residual Network (DCRNet) to cope with the data skewness problem. The results demonstrated that the proposed approach improves the classification performance over the UQ-SNP I3A Task-3 mitotic cell dataset with the advantage of using an end-to-end CNN classifier.
KW - Computer-aided diagnosis (CADs)
KW - Convolutional neural network (CNN)
KW - HEp2 mitotic patterns
KW - One-dimensional GAN
KW - Oversampling
UR - http://www.scopus.com/inward/record.url?scp=85129188810&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85129188810&partnerID=8YFLogxK
U2 - 10.1109/LifeTech53646.2022.9754940
DO - 10.1109/LifeTech53646.2022.9754940
M3 - Conference contribution
AN - SCOPUS:85129188810
T3 - LifeTech 2022 - 2022 IEEE 4th Global Conference on Life Sciences and Technologies
SP - 172
EP - 174
BT - LifeTech 2022 - 2022 IEEE 4th Global Conference on Life Sciences and Technologies
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 4th IEEE Global Conference on Life Sciences and Technologies, LifeTech 2022
Y2 - 7 March 2022 through 9 March 2022
ER -
