Abstract:
Ovarian cancer presents a significant challenge to global healthcare, demanding accurate
diagnosis and specialized treatment plans. Leveraging advancements in deep learning,
particularly in medical imaging, offers a promising approach to enhance subtype
classification. This study introduces a robust deep learning framework for precise
categorization of ovarian cancer subtypes from image data. Driven by the urgent need to
improve diagnostic accuracy, our system integrates deep learning methods with
convolutional neural networks (CNNs) to deliver reliable classification outcomes. Through
preprocessing of image data and utilization of pre-trained CNN models for feature
extraction, complex patterns corresponding to various ovarian cancer subtypes are
identified. Classification is further enhanced through fine-tuning and ensemble learning
techniques to maximize performance. The proposed framework addresses critical gaps in
current diagnostic methods, providing clinicians with a rapid and accurate tool for subtype
classification. Rigorous experimentation and validation aim to develop a system capable
of generalizing across diverse patient populations, ultimately improving patient outcomes
in ovarian cancer treatment. Experimental results demonstrate that our proposed model
achieves a significant accuracy of 96%, along with 96% precision, recall, and F1-score,
and a 97% area under the ROC curve (AUC-ROC) score. While our model's accuracy
surpasses that of traditional transfer learning models such as VGG16, VGG19, Xception,
and EfficientNetB4, further refinements are essential to maximize its potential in clinical
applications. This study underscores the importance of advancing medical research through
data-driven approaches and highlights the critical role of ongoing progress in healthcare.
