Context-specific selection of algorithms for recursive feature tracking in endoscopic image using a new methodology.

In minimally invasive surgery, the tracking of deformable tissue is a critical component for image-guided applications. Deformation of the tissue can be recovered by tracking features using tissue surface information (texture, color,...). Recent work in this field has shown success in acquiring tissue motion. However, the performance evaluation of detection and tracking algorithms on such images are still difficult and are not standardized. This is mainly due to the lack of ground truth data on real data. Moreover, in order to avoid supplementary techniques to remove outliers, no quantitative work has been undertaken to evaluate the benefit of a pre-process based on image filtering, which can improve feature tracking robustness. In this paper, we propose a methodology to validate detection and feature tracking algorithms, using a trick based on forward-backward tracking that provides an artificial ground truth data. We describe a clear and complete methodology to evaluate and compare different detection and tracking algorithms. In addition, we extend our framework to propose a strategy to identify the best combinations from a set of detector, tracker and pre-process algorithms, according to the live intra-operative data. Experimental results have been performed on in vivo datasets and show that pre-process can have a strong influence on tracking performance and that our strategy to find the best combinations is relevant for a reasonable computation cost.

Extraction of specific parameters for skin tumour classification.

In this paper, a methodological approach to the classification of tumour skin lesions in dermoscopy images is presented. Melanomas are the most malignant skin tumours. They grow in melanocytes, the cells responsible for pigmentation. This type of cancer is increasing rapidly; its related mortality rate is increasing more modestly, and inversely proportional to the thickness of the tumour. The mortality rate can be decreased by earlier detection of suspicious lesions and better prevention. Using skin tumour features such as colour, symmetry and border regularity, an attempt is made to determine if the skin tumour is a melanoma or a benign tumour. In this work, we are interested in extracting specific attributes which can be used for computer-aided diagnosis of melanoma, especially among general practitioners. In the first step, we eliminate surrounding hair in order to eliminate the residual noise. In the second step, an automatic segmentation is applied to the image of the skin tumour. This method reduces a colour image into an intensity image and approximately segments the image by intensity thresholding. Then, it refines the segmentation using the image edges, which are used to localize the boundary in that area of the skin. This step is essential to characterize the shape of the lesion and also to locate the tumour for analysis. Then, a sequences of transformations is applied to the image to measure a set of attributes (A: asymmetry, B: border, C: colour and D: diameter) which contain sufficient information to differentiate a melanoma from benign lesions. Finally, the various signs of specific lesion (ABCD) are provided to an artificial neural network to differentiate between malignant tumours and benign lesions.