Performance Analysis of Ovarian Cancer Detection and Classification for Microarray Gene Data.

Area of Science:

• Biomedical Engineering
• Genomics
• Computational Biology

Background:

• Ovarian cancer is a leading gynecologic malignancy, necessitating improved diagnostic and prognostic tools.
• Microarray data for ovarian cancer contains numerous genes, posing computational challenges for accurate classification.
• Dimensionality reduction is crucial for effective gene selection in high-dimensional microarray datasets.

Purpose of the Study:

• To investigate feature extraction methods for dimensionality reduction in ovarian cancer microarray data.
• To identify critical genes for improved classification of normal versus abnormal samples.
• To evaluate the performance of different classifiers on selected features.

Main Methods:

• Applied Analysis of Variance (ANOVA) for initial gene selection.
• Utilized clustering-based (Fuzzy C Means) and transform-based (Softmax Discriminant Algorithm, Hilbert Transform, Fast Fourier Transform, Discrete Cosine Transform) feature extraction techniques.
• Employed six distinct classifiers to differentiate between normal and abnormal samples based on extracted features.

Main Results:

• The NLR classifier achieved the highest accuracy (92%) when using Softmax Discriminant Algorithm (SDA) features.
• KNN classifier showed the lowest accuracy (55%) with SDA, Hilbert, and DCT features.
• Using correlation distance for feature selection, the GMM classifier attained the highest accuracy (88%), while NLR achieved the lowest (53%).

Conclusions:

• Feature extraction techniques significantly impact the accuracy of ovarian cancer classification from microarray data.
• The Softmax Discriminant Algorithm (SDA) combined with the NLR classifier demonstrates high potential for accurate ovarian cancer diagnosis.
• Further research into feature selection and classification algorithms is warranted for robust clinical application.