WOAENet: a whale optimization-guided ensemble deep learning with soft voting for uterine cancer diagnosis based on MRI images

Objectives: Uterine cancer originates from the cells lining the uterus and can develop through abnormal cell growth, potentially leading to damage in surrounding tissues and the formation of precancerous cells. Early detection significantly improves prognosis. Despite advancements in deep learning-based diagnostic methods, challenges remain, including the dependence on expert input and the need for more accurate classification models. This study aims to address these limitations by proposing a novel and efficient methodology for diagnosing uterine cancer using an integrated deep learning pipeline optimized through a nature-inspired algorithm. Methods: This study introduces the Whale Optimization Algorithm-based Ensemble Network (WOAENet), a deep learning pipeline that classifies uterine MRI into three classes: malignant, benign, and normal. The Whale Optimization Algorithm (WOA) is used to fine-tune the hyperparameters of three deep learning models: MobileNetV2, DenseNet121, and a lightweight vision model (LVM). Each model is trained with its optimized settings, and its outputs are combined using a Soft Voting Ensemble method that calculates the average of the predicted probabilities to arrive at the final classification. Results: The WOAENet framework was evaluated using a uterine cancer MRI dataset obtained from King Abdullah University Hospital. Our proposed model outperformed standard pre-trained models across several performance metrics. It achieved an accuracy of 88.57%, a specificity of 94.29%, and an F1 score of 88.54%, indicating superior performance in diagnosing uterine cancer. Conclusion: WOAENet demonstrates a high level of accuracy and reliability in classifying uterine MRI images, marking a significant advancement by utilizing a novel dataset. The findings support the potential of AI-driven approaches in enhancing the diagnosis and treatment of gynecological conditions, paving the way for more accessible and accurate clinical tools.