ANN based dementia diagnosis using DCT for brain MR image compression

The paper proposes use of an innovative method for automated multiclass diagnosis of Dementia, based on classification of magnetic resonance images (MRI) of human brain. 1D histogram signal is obtained from 2D MR images of brain and then further compression is done using discrete cosine transform. The proposed method uses first few DCT coefficients as features for the ANN classification. The features hence derived are used to train a neural network based four class classifier, which can automatically infer whether the MR image belongs to a normal brain or to a person suffering from Alzheimer´s disease or Mild Alzheimer´s disease or Huntington´s Disease. An excellent classification rate of 100% is achieved for a set of benchmark MR brain images from the Whole brain atlas database at http://www.med.harvard.edu/AANLIB/nav.htm.