Morphological segmenting and neighborhood pixel-based locality preserving projection on brain fMRI dataset for semantic feature extraction: an affective computing study

Two specific chemical receptive fields of brain, namely the amygdala and the orbital-frontal cortex, are related to valence and arousal in medical experiments. Functional magnetic resonance imaging (fMRI), which is a noninvasive, repeatable, and atomical tool for medical imaging in clinic system, was widely used in affective computing; however, it faces its dataset processing difficulty for dimensional reduction as well as for decreasing the computational complexity. In addition, features extraction from those de-dimensionality datasets is a challenging issue. The current work solved the de-dimensionality issue by using some preprocessing algorithms including clustering, morphological segmenting, and locality preserving projection. In order to keep useful information in fMRI dataset for reduction process, improved neighborhood pixel-based locality preserving projection (NP-LPP) algorithm was addressed and continuously for feature extraction operating using Otsu weighted sum of histogram. Furthermore, a modified covariance power spectral density (MC-PSD) separately in an fMRI Valence–Arousal experiments was measured. The results were analyzed and compared with affective norms English words system. The experiments established that the proposed methods of NP-LPP effectively simplified high complexity of fMRI, and Otsu weighted sum of histogram exhibited superior performance for features extraction compared to the MC-PSD through the calculation root mean standard error. The current proposed method provided a potential application and promising research direction on human semantic retrieval through medical imaging dataset.