Lesion Border Detection of Skin Cancer Images Using Deep Fully Convolutional Neural Network with Customized Weights.

Deep learning techniques have been widely employed in semantic segmentation problems, especially in medical image analysis, for understanding image patterns. Skin cancer is a life-threatening problem, whereas timely detection can prevent and reduce the mortality rate. The aim is to segment the lesion area from the skin cancer image to help experts in the process of deeply understanding tissues and cancer cells' formation. Thus, we proposed an improved fully convolutional neural network (FCNN) architecture for lesion segmentation in dermoscopic skin cancer images. The FCNN network consists of multiple feature extraction layers forming a deep framework to obtain a larger vision for generating pixel labels. The novelty of the network lies in the way layers are stacked and the generation of customized weights in each convolutional layer to produce a full resolution feature map. The proposed model was compared with the top four winners of the International Skin Imaging Collaboration (ISIC) challenge using evaluation metrics such as accuracy, Jaccard index, and dice co-efficient. It outperformed the given state-of-the-art methods with higher values of the accuracy and Jaccard index.

Integration of vibro-acoustography imaging modality with the traditional mammography.

Vibro-acoustography (VA) is a new imaging modality that has been applied to both medical and industrial imaging. Integrating unique diagnostic information of VA with other medical imaging is one of our research interests. In this work, we establish correspondence between the VA images and traditional X-ray mammogram by adopting a flexible control-point selection technique for image registration. A modified second-order polynomial, which simply leads to a scale/rotation/translation invariant registration, was used. The results of registration were used to spatially transform the breast VA images to map with the X-ray mammography with a registration error of less than 1.65 mm. The fused image is defined as a linear integration of the VA and X-ray images. Moreover, a color-based fusion technique was employed to integrate the images for better visualization of structural information.

The comparison of different feed forward neural network architectures for ECG signal diagnosis.

The electrocardiograms (ECGs) record the electrical activity of the heart and are used to diagnose many heart disorders. This paper proposes a two-stage feed forward neural network for ECG signal classification. The research is aimed at the design of an intelligent ECG diagnosis tool that can recognise heart abnormalities while reducing the complexity, cost, and response time of the system. A number of neural network architectures are designed and compared for their ability to classify six different heart conditions. Two network architectures based on one stage and two stage feed forward neural networks are chosen for this investigation. The training and testing ECG signals are obtained from MIT-BIH database. The network inputs are comprised of 12 ECG features and 13 compressed components of each heart beat signal. The performance of the different modules as well as the efficiency of the whole system is presented. Among different architectures, a proposed multi-stage network named NET_BST possesses the highest recognition rate of around 93%. Therefore, this network proves to be a suitable candidate in ECG signal diagnosis systems.

Extracting effective features of SEMG using continuous wavelet transform.

To date various signal processing techniques have been applied to surface electromyography (SEMG) for feature extraction and signal classification. Compared with traditional analysis methods which have been used in previous application, continuous wavelet transform (CWT) enhances the SEMG features more effectively. This paper presents methods of analysing SEMG signals using CWT and LabVIEW for extracting accurate patterns of the SEMG signals. We used the scalogram and frequency-time based spectrum to plot the power of the wavelet transform and enhance the diagnosis features of the signal. As a result, clinical interpretation of SEMG can be improved by extracting time-based information as well as scales, which can be converted to frequencies. Using the extracted features of the dominant frequencies of the wavelet transform and the related scales, we were able to train and validate an artificial neural network (ANN) for SEMG classification.

Facial expression analysis for estimating patient's emotional states in RPMS.

Currently, a range of remote patient monitoring systems (RPMS) are being developed to care for patients at home rather than in the costly hospital environment. These systems allow remote monitoring by health professionals with minimum medical intervention to take place. However, they are still not as effective as one-on-one human interaction. The face and its features can convey patient cognitive and emotional states faster than electrical signals and facial expression can be considered as one of the most powerful features of RPMS. We present image pre-processing and enhancement techniques for face recognition applications. In particular, the project is aimed to improve the performance of RPMS, taking into account the cognitive and emotional state of patients by developing a more human like RPMS. The techniques use the value of grey scale of the images and extract efficient facial features. The extracted information is fed into input layer of an artificial neural network for face identification. On the other hand, the colour images are used by the recognition algorithm to eliminate nonskin coloured background and reduce further processing time. A data base of real images is used for testing the algorithms.