Probabilistic guided polycystic ovary syndrome recognition using learned quality kernel

Abstract Image recognition aims to automatically search special objects in an image, such as human faces, vehicles, or buildings. In medical research, image recognition technique can also be applied for disease diagnosis and disease classification. Aiming at disadvantages of traditional methods in polycystic ovary syndrome (PCOS) recognition, we propose a probabilistic model for disease recognition using a deeply-learned image quality kernel. Specifically, we first segment training images into several equal-size grids for better cues discovery. Then, each grid within an image is quantitatively represented by a quality score according to grayscale and texture features. In this way, each image can be represented by a score matrix. Then, we leverage statistic based method to generate a long feature vector according to the score matrix. Afterward, we propose a probabilistic model to learn the distribution of obtained feature vector, which will be further fed into a SVM kernel for PCOS recognition. Experimental results show the effectiveness of our proposed method.