Ovarian Tumor Detection Using Machine Learning Algorithms and Local Texture Features
Nidaa Ghalib Ali, Farsad Zamani Boroujeni, Sahar Adill Kadum, Mehdi Akbari Kopayei, Mahdi Mosleh · Qubahan Academic Journal · 2025
Ovarian cancer (OC) is considered the fifth leading cause of death among women globally, and early detection of ovarian cancer symptoms will be vital for ovarian cancer treatment. Its detection via image-based clinical diagnosis is often prone to misclassification. Therefore, this study proposes using machine learning methods to reduce such errors and deliver faster and more accurate results. Two ovarian tumor datasets consisting of ultrasound images were analyzed using nine machine learning classification algorithms. To reduce the extracted features derived from four pre-trained convolutional networks and five texture-based features to fewer than 50. A hierarchical feature selection method was proposed, combining the ReliefF filter algorithm in the first stage and ten metaheuristic algorithms in the second stage. The results showed that ensemble algorithms, such as LightGBM, achieved a high accuracy of over 95% in diagnosing various types of ovarian tumors using both 2D and 3D ultrasound images. Among the feature selection approaches, the combination of ReliefF and Quantum Approximate Neighbourhood Analysis (QANA) yielded the best performance. Experimental findings on real datasets show that the suggested method not only preserves data confidentiality but also yields excellent performance in early and precise detection of OC.