Diana Veiga

Using a multi-agent system approach for microaneurysm detection in fundus images

OBJECTIVE Microaneurysms represent the first sign of diabetic retinopathy, and their detection is fundamental for the prevention of vision impairment. Despite several research attempts to develop an automated system to detect microaneurysms in fundus images, none has shown the level of performance required for clinical practice. We propose a new approach, based on a multi-agent system model, for microaneurysm segmentation. METHODS AND MATERIALS A multi-agent based approach, preceded by a preprocessing phase to allow construction of the environment in which agents are situated and interact, is presented. The proposed method is applied to two available online datasets and results are compared to other previously described approaches. RESULTS Microaneurysm segmentation emerges from agent interaction. The final score of the proposed approach was 0.240 in the Retinopathy Online Challenge. CONCLUSIONS We achieved competitive results, primarily in detecting microaneurysms close to vessels, compared to more conventional algorithms. Despite these results not being optimum, they are encouraging and reveal that some improvements may be made.

Quality evaluation of digital fundus images through combined measures

Abstract. The evaluation of image quality is an important step before an automatic analysis of retinal images. Several conditions can impair the acquisition of a good image, and minimum image quality requirements should be present to ensure that an automatic or semiautomatic system provides an accurate diagnosis. A method to classify fundus images as low or good quality is presented. The method starts with the detection of regions of uneven illumination and evaluates if the segmented noise masks affect a clinically relevant area (around the macula). Afterwards, focus is evaluated through a fuzzy classifier. An input vector is created extracting three focus features. The system was validated in a large dataset (1454 fundus images), obtained from an online database and an eye clinic and compared with the ratings of three observers. The system performance was close to optimal with an area under the receiver operating characteristic curve of 0.9943.

Automatic microaneurysm detection using laws texture masks and support vector machines

AbstractMicroaneurysms are one of the first clinical signs of diabetic retinopathy, and their detection is crucial for an early diagnosis of the disease. In this contribution, an approach based on Laws texture features to detect microaneurysms is presented. The proposed algorithm uses support vector machines (SVM) in two classification phases. The first SVM performs a pixel-based classification to find the microaneurysm candidates. The second SVM performs an object classification, aiming to reduce the false detections of the first classifier. The algorithm performance was evaluated on three public data-sets. The results show sensitivities of 62, 66 and 32%, for an average number of 10 false positives per image, in LaTIM, e-ophtha and ROC databases, respectively. Laws texture masks proved to be good features to detect microaneurysm candidates. Moreover, the second classification phase achieved to reducing the false detections.

Tracking of the anterior mitral leaflet in echocardiographic sequences using active contours

Echocardiography assessment of cardiac valves plays a vital role in the diagnosis of rheumatic heart disease. In the vast majority of cases, the mitral valve gets affected, leading to the thickening of its leaflets that may result in the fusion of their tips. This changes the appearance and reduces the mobility of the leaflets, which also reduce the heart efficiency. Quantifying such parameters provides diagnostic insight. To achieve that, the first step is to identify and then track fast moving leaflets. This work is focused on Anterior Mitral Leaflet (AML) tracking. Open ended active contours are employed in this work by removing its boundary conditions. The external and internal energy of the contour is modified that extend the capture range, improve snake energy and encourages the leftmost end point of the contour to converge on the moving tip of the AML. Results show that contour points are tracked accurately with an average error of 4.9 pixels and a standard deviation of 2.1 pixels in 9 fully annotated normal sequences of real children clinical assessments.

Real-time Anterior Mitral Leaflet Tracking using Morphological Operators and Active Contours.

The mitral valve plays a vital role in our circulatory system. To study its functionality, it is important to measure clinically relevant parameters, such as its thickness, mobility and shape. Since manual segmentation is impractical, time consuming and requires expert knowledge, an automatic segmentation tool can have a significant clinical impact, providing objective measures to clinicians for understanding the morphology and behaviour of the mitral valve. In this work, a real time tracking method has been proposed for ultrasound videos obtained with the Parasternal Long Axis view. The algorithm is semi-automatic, assumes manual Anterior Mitral Leaflet segmentation in the first frame and then it uses mathematical morphology algorithms to obtain tracking results, further refined by localized active contours during the whole cardiac cycle. Finally, the medial axis is extracted for a quantitative analysis. Results show that the algorithm can segment 1137 frames extracted from 9 fully annotated sequences of the real clinical video data in only 0.89 sec/frame, with an average error of 5 pixels. Furthermore, the algorithms exhibited robust tracking performance in the most difficult situations, which are large frame-to-frame displacements.

Segmentation of the metacarpus and phalange in musculoskeletal ultrasound images using local active contours

This work presents a method for the automatic segmentation of metacarpus and phalange bones in ultrasound images of the second metacarpophalangeal joint (MCPJ) using Active Contours. The MCPJ is known to be the one of the first structures to be affected by rheumatic diseases like rheumatoid arthritis. The early detection and follow-up of this disease is important to prevent irreversible damage of the joints, which occurs continuously and faster if no treatment is used. To our knowledge, there is no automatic system to quantify the extension of the lesions resulting from rheumatic activity. The objective of this work is to identify the metacarpus and the phalange bones using local active contours. To our knowledge, there is no well established method for this problem and this technique has not been used yet in these structures. Results proved that the automatic segmentation is possible with an error of 3 pixels for a confidence of 80%.

Focus evaluation approach for retinal images

Digital fundus photographs are often used to provide clinical diagnostic information about several pathologies such as diabetes, glaucoma, macular degeneration and vascular and neurologic disorders. To allow a precise analysis, digital fundus image quality should be assessed to evaluate if minimum requirements are present. Focus is one of the causes of low image quality. This paper describes a method that automatically classifies fundus images as focused or defocused. Various focus measures described in literature were tested and included in a feature vector for the classification step. A neural network classifier was used. HEI-MED and MESSIDOR image sets were utilized in the training and testing phase, respectively. All images were correctly classified by the proposed algorithm.

Tracking Anterior Mitral Leaflet in Echocardiographic Videos Using Morphological Operators and Active Contours

Rheumatic heart disease is the result of damage to the heart valves, more often the mitral valve. The heart valves leaflets get inflamed, scarred and stretched which interrupts the normal blood flow, resulting into serious health condition. Measuring and quantifying clinically relevant features, like thickness, mobility and shape can help to analyze the functionality of the valve, identify early cases of disease and reduce the disease burden. To obtain these features, the first step is to automatically delineate the relevant structures, such as the anterior mitral valve leaflet, throughout the echocardiographic video. In this work, we proposed a near real time method to track the anterior mitral leaflet in ultrasound videos using the parasternal long axis view. The method is semi-automatic, requiring a manual delineation of the anterior mitral leaflet in the first frame of the video. The method uses mathematical morphological techniques to obtain the rough boundaries of the leaflet and are further refined by the localized active contour framework. The mobility of the leaflet was also obtained, providing us the base to analyze the functionality of the valve (opening and closing). The algorithm was tested on 67 videos with 6432 frames. It outperformed with respect to the time consumption (0.4 s/frame), with the extended modified Hausdorff distance error of 3.7 pixels and the improved tracking performance (less failure).

A Preliminary Approach for the Segmentation of Mitral Valve Regurgitation Jet in Doppler Ecocardiography Images.

Rheumatic Fever and Rheumatic Heart Disease remain a major burden among children in developing countries. Echocardiography with colour flow Doppler is key to early diagnosis. However, the technique requires time and experienced operators, which are scarce resources in the affected areas. Automatic segmentation of colour Doppler regurgitation jets could, potentially, reduce the cost of screening, and spread diagnostic accessibility for a larger number of patients. Ultrasound processing is very challenging due to speckle noise and similarity of representation of all kinds of tissue. Region-based active contours are suitable tools for the segmentation in cases of intensity heterogeneities, which makes them interesting algorithms for left atrium segmentation. HSV colour space describes colour in terms of hues and saturation, which may facilitate the translation of medical interpretation of the Doppler pseudo-colour into mathematical expression for colour segmentation. A total of 979 frames from 20 sequences were manually annotated and used to validate the proposed pipeline. Overall, the results for colour pattern segmentation are promising (sensitivity=0.91 false detection rate=0.10), but further developments are required for the atrium segmentation (sensitivity=0.80, false detection rate=0.28).

Automatic Arteriovenous Nicking Identification by Color Fundus Images Analysis

Retinal arteriovenous nicking (AVN) assessment has been considered a very important indicator of cardiovascular and cerebrovascular diseases. A computerized method to infer the AVN presence in retinal images could increase the reproducibility and accuracy of this analysis that for now has been done by ophthalmologists in a subjective and qualitative manner. Therefore, a new approach is proposed for the AVN assessment in color fundus images. First the algorithm segments the blood vessels by means of a multi-scale line detector. The arteriovenous cross points are then detected and classified as AVN presence or absence with an SVM. The proposed approach is clearly efficient in separating normal cases from the evident or severe AVN cases.