TY - GEN
T1 - Multi-scale Contrastive Learning for Gastroenteroscopy Classification
AU - Li, Dan
AU - Li, Xuechen
AU - Peng, Zhibin
AU - Chen, Wenting
AU - Shen, Linlin
AU - Wu, Guangyao
N1 - Publisher Copyright:
© 2023 IEEE.
PY - 2023
Y1 - 2023
N2 - In gastroenteroscopy image analysis, numerous CADs demonstrate that deep learning aids doctors' diagnosis. The shapes and sizes of the lesions are varied. And in the clinic, the dataset appears to be data imbalanced. However, existing methods directly classify by texture and ignore lesions with various shapes and sizes. To address the issue above, we propose a deep neural network, which consists of multi-scale feature extraction, contrastive feature learning and a multi-scale feature fusion module. We train the contrastive feature learning module and multi-scale feature fusion module simultaneously to alleviate the issue of data distribution differences. Thus, the proposed network can better identify various categories. Extensive experiments on the Hyper Kvasir dataset show that the proposed Hybrid-M2CL outperforms the benchmark proposed by the dataset with 5.0% Macro Precision, 3.3% Macro Recall, 3.4% Macro F1-score, 3.3% Micro Precision, 3.6% MCC. In addition, it outperforms the SOTA by 1.1% Macro F1-score, 2.6% MCC, and 2.0% B-ACC.
AB - In gastroenteroscopy image analysis, numerous CADs demonstrate that deep learning aids doctors' diagnosis. The shapes and sizes of the lesions are varied. And in the clinic, the dataset appears to be data imbalanced. However, existing methods directly classify by texture and ignore lesions with various shapes and sizes. To address the issue above, we propose a deep neural network, which consists of multi-scale feature extraction, contrastive feature learning and a multi-scale feature fusion module. We train the contrastive feature learning module and multi-scale feature fusion module simultaneously to alleviate the issue of data distribution differences. Thus, the proposed network can better identify various categories. Extensive experiments on the Hyper Kvasir dataset show that the proposed Hybrid-M2CL outperforms the benchmark proposed by the dataset with 5.0% Macro Precision, 3.3% Macro Recall, 3.4% Macro F1-score, 3.3% Micro Precision, 3.6% MCC. In addition, it outperforms the SOTA by 1.1% Macro F1-score, 2.6% MCC, and 2.0% B-ACC.
KW - contrastive learning
KW - gastroenterology classification
KW - multi-scale
KW - supervised learning
UR - http://www.scopus.com/inward/record.url?scp=85166470240&partnerID=8YFLogxK
U2 - 10.1109/CBMS58004.2023.00331
DO - 10.1109/CBMS58004.2023.00331
M3 - Conference contribution
AN - SCOPUS:85166470240
T3 - Proceedings - IEEE Symposium on Computer-Based Medical Systems
SP - 852
EP - 858
BT - Proceedings - 2023 IEEE 36th International Symposium on Computer-Based Medical Systems, CBMS 2023
A2 - Sicilia, Rosa
A2 - Kane, Bridget
A2 - Almeida, Joao Rafael
A2 - Spiliopoulou, Myra
A2 - Andrades, Jose Alberto Benitez
A2 - Placidi, Giuseppe
A2 - Gonzalez, Alejandro Rodriguez
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 36th IEEE International Symposium on Computer-Based Medical Systems, CBMS 2023
Y2 - 22 June 2023 through 24 June 2023
ER -
