Classification of meningiomas using discriminant wavelet packets and learning vector quantization
H. Qureshi, Nasir Rajpoot, Khalid Masood, Volkmar Hans · Warwick Research Archive Portal (University of Warwick) · 2006
Abstract — This paper presents a novel texture-based algorithm for detecting certain kinds of meningiomas in images of neurosurgical resections. The algorithm employs Discriminant Wavelet Packet Transform (DWPT) and Learning Vector Quantization (LVQ). The adaptive DWPT of a test image is computed by maximizing the discrimination power of subbands during the basis selection process for the training images. The discrimination power of a subband is computed using the Hellinger distance between pseudo probability density functions of the subband. Statistical features are then obtained for each of the top few most discriminant subbands and finally LVQ is trained and subsequently used for detection. The proposed methodology produces promising results for the multiple class meningioma classification problem with near-perfect results for detection of certain kind of meningiomas. I.