TY - JOUR
T1 - Advancing Ovarian Cancer Diagnosis through Deep Learning and eXplainable AI
T2 - A Multiclassification Approach
AU - Radhakrishnan, Meera
AU - Sampathila, Niranjana
AU - Muralikrishna, H.
AU - Swathi, K. S.
N1 - Publisher Copyright:
© 2013 IEEE.
PY - 2024
Y1 - 2024
N2 - Ovarian cancer is a dangerous gynaecological malignancy, and the presence of many subtypes causes significant diagnostic difficulties. In general, the high accuracy of classification results in adequate prognosis and effectiveness of treatment. This work aims at the development of a Deep Learning (DL) approach for subtypes of ovarian cancer multiclassification, which tries to solve the problem of the creation of precise and reliable diagnostic methods. In the work, we have used and explored various DL models such as MobileNetV2, VGG19, ResNet18, ResNeXt, Xception, EfficientNet, and InceptionV3 to perform the classification task. Further, we used the state-of-the-art eXplainable Artificial Intelligence methods, including integrated gradient, saliency map, Grad-CAM, and DeepLift, to improve model interpretability. From our experiments, we inferred that the highest accuracy was achieved by InceptionV3, with a value of 97.96%. XAI techniques incorporated provide transparent insights into the model's operations during the decision-making process, thus increasing the level of trust and clinical usability. The proposed DL approach, by leveraging InceptionV3 as its top performer, has convincingly demonstrated the potential of AI to revolutionize the diagnosis of ovarian cancer through a high level of accuracy in subtype classification. XAI techniques integrated allow transparency support for the model and further enable its clinical adoption. All of these developments have significant potential for improved patient outcomes within the scope of personalized medicine in ovarian cancer treatment.
AB - Ovarian cancer is a dangerous gynaecological malignancy, and the presence of many subtypes causes significant diagnostic difficulties. In general, the high accuracy of classification results in adequate prognosis and effectiveness of treatment. This work aims at the development of a Deep Learning (DL) approach for subtypes of ovarian cancer multiclassification, which tries to solve the problem of the creation of precise and reliable diagnostic methods. In the work, we have used and explored various DL models such as MobileNetV2, VGG19, ResNet18, ResNeXt, Xception, EfficientNet, and InceptionV3 to perform the classification task. Further, we used the state-of-the-art eXplainable Artificial Intelligence methods, including integrated gradient, saliency map, Grad-CAM, and DeepLift, to improve model interpretability. From our experiments, we inferred that the highest accuracy was achieved by InceptionV3, with a value of 97.96%. XAI techniques incorporated provide transparent insights into the model's operations during the decision-making process, thus increasing the level of trust and clinical usability. The proposed DL approach, by leveraging InceptionV3 as its top performer, has convincingly demonstrated the potential of AI to revolutionize the diagnosis of ovarian cancer through a high level of accuracy in subtype classification. XAI techniques integrated allow transparency support for the model and further enable its clinical adoption. All of these developments have significant potential for improved patient outcomes within the scope of personalized medicine in ovarian cancer treatment.
UR - http://www.scopus.com/inward/record.url?scp=85201779169&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85201779169&partnerID=8YFLogxK
U2 - 10.1109/ACCESS.2024.3448219
DO - 10.1109/ACCESS.2024.3448219
M3 - Article
AN - SCOPUS:85201779169
SN - 2169-3536
VL - 12
SP - 116968
EP - 116986
JO - IEEE Access
JF - IEEE Access
ER -