Search

Featured Application: This paper presents a novel deep learning system for the detection and diagnosis of diabetic retinopathy using optical coherence tomography images. Diabetic retinopathy (DR) is a major public health problem and the leading cause of vision loss in the working age population. This paper presents a novel deep learning system for the detection and diagnosis of DR using optical coherence tomography (OCT) images. The input for this system is three-channel local and global information from OCT images. The local high-level information is represented by the thickness channel and the reflectivity channel. The global low-level information is represented by the grey-level OCT original image. The deep learning system processes the three-channel input to produce the final DR diagnoses. Experimental results on 200 OCT images, augmented to 800 images, which are collected by the University of Louisville, show high system performance related to other competing methods. Moreover, 10-fold and leave-one-subject-out (LOSO) experiments are performed to confirm how significant using the fused images is in improving the performance of the diagnoses, by investigating four different CNN architectures. All of the four architectures achieve acceptable performance and confirm a significant performance improvement using the fused images. Using LOSO, the best network performance has improved from 90.1 ± 2% using only the grey level dataset to 97.7 ± 0.5% using the proposed fused dataset. These results confirm the promise of using the proposed system for the detection of DR using OCT images. [ABSTRACT FROM AUTHOR]

Diabetic retinopathy (DR) is a major public health problem and the leading cause of vision loss in the working age population. This paper presents a novel deep learning system for the detection and diagnosis of DR using optical coherence tomography (OCT) images. The input for this system is three-channel local and global information from OCT images. The local high-level information is represented by the thickness channel and the reflectivity channel. The global low-level information is represented by the grey-level OCT original image. The deep learning system processes the three-channel input to produce the final DR diagnoses. Experimental results on 200 OCT images, augmented to 800 images, which are collected by the University of Louisville, show high system performance related to other competing methods. Moreover, 10-fold and leave-one-subject-out (LOSO) experiments are performed to confirm how significant using the fused images is in improving the performance of the diagnoses, by investigating four different CNN architectures. All of the four architectures achieve acceptable performance and confirm a significant performance improvement using the fused images. Using LOSO, the best network performance has improved from 90.1 ± 2% using only the grey level dataset to 97.7 ± 0.5% using the proposed fused dataset. These results confirm the promise of using the proposed system for the detection of DR using OCT images.

Highlights: [•] We propose a generative approach to transfer semantic label from a training image to test image. [•] We propose a new fast nonparametric approach to scene parsing. [•] Our approach does not require patch matching in the training and test phases. [•] It incorporates object level information and contextual information. [•] It outperforms state-of-the-art nonparametric approaches. [Copyright &y& Elsevier]

Recent advances in speech technologies have produced new tools that can be used to improve the performance and flexibility of speaker recognition. While there are few degrees of freedom or alternative methods when using fingerprint or iris identification techniques, speech offers much more flexibility and different levels to perform recognition: the system can force the user to speak in a particular manner, different for each attempt to enter. Also, with voice input, the system has other degrees of freedom, such as the use of knowledge/codes that only the user knows, or dialectical/semantical traits that are difficult to forge. This paper offers an overview of the state-of-the-art in speaker recognition, with special emphasis on the pros and cons, and the current research lines. The current research lines include improved classification systems, and the use of high level information by means of probabilistic grammars. In conclusion, speaker recognition is far away from being a technology where all the possibilities have already been explored. [ABSTRACT FROM PUBLISHER]