Lama Seoud

Red Lesion Detection Using Dynamic Shape Features for Diabetic Retinopathy Screening

The development of an automatic telemedicine system for computer-aided screening and grading of diabetic retinopathy depends on reliable detection of retinal lesions in fundus images. In this paper, a novel method for automatic detection of both microaneurysms and hemorrhages in color fundus images is described and validated. The main contribution is a new set of shape features, called Dynamic Shape Features, that do not require precise segmentation of the regions to be classified. These features represent the evolution of the shape during image flooding and allow to discriminate between lesions and vessel segments. The method is validated per-lesion and per-image using six databases, four of which are publicly available. It proves to be robust with respect to variability in image resolution, quality and acquisition system. On the Retinopathy Online Challenges database, the method achieves a FROC score of 0.420 which ranks it fourth. On the Messidor database, when detecting images with diabetic retinopathy, the proposed method achieves an area under the ROC curve of 0.899, comparable to the score of human experts, and it outperforms state-of-the-art approaches.

Multilevel analysis of trunk surface measurements for noninvasive assessment of scoliosis deformities.

Study Design. Reliability study. Objectives. To assess between-acquisition reliability of new multilevel trunk cross sections measurements, in order to define what is a real change when comparing 2 trunk surface acquisitions of a same patient, before and after surgery or throughout the clinical monitoring. Summary of Background Data. Several cross-sectional surface measurements have been proposed in the literature for noninvasive assessment of trunk deformity in patients with adolescent idiopathic scoliosis (AIS). However, only the maximum values along the trunk are evaluated and used for monitoring progression and assessing treatment outcome. Methods. Back surface rotation (BSR), trunk rotation (TR), and coronal and sagittal trunk deviation are computed on 300 cross sections of the trunk. Each set of 300 measures is represented as a single functional data, using a set of basis functions. To evaluate between-acquisition variability at all trunk levels, a test-retest reliability study is conducted on 35 patients with AIS. A functional correlation analysis is also carried out to evaluate any redundancy between the measurements. Results. Each set of 300 measures was successfully described using only 10 basis functions. The test-retest reliability of the functional measurements is good to very good all over the trunk, except above the shoulders level. The typical errors of measurement are between 1.20° and 2.2° for the rotational measures and between 2 and 6 mm for deviation measures. There is a very strong correlation between BSR and TR all over the trunk, a moderate correlation between coronal trunk deviation and both BSR and TR, and no correlation between sagittal trunk deviation and any other measurement. Conclusion. This novel representation of trunk surface measurements allows for a global assessment of trunk surface deformity. Multilevel trunk measurements provide a broader perspective of the trunk deformity and allow a reliable multilevel monitoring during clinical follow-up of patients with AIS and a reliable assessment of the esthetic outcome after surgery.

Prediction of scoliosis curve type based on the analysis of trunk surface topography

Scoliosis treatment strategy is generally chosen according to the severity and type of the spinal curve. Currently, the curve type is determined from X-rays whose acquisition can be harmful for the patient. We propose in this paper a system that can predict the scoliosis curve type based on the analysis of the surface of the trunk. The latter is acquired and reconstructed in 3D using a non invasive multi-head digitizing system. The deformity is described by the back surface rotation, measured on several cross-sections of the trunk. A classifier composed of three support vector machines was trained and tested using the data of 97 patients with scoliosis. A prediction rate of 72.2% was obtained, showing that the use of the trunk surface for a high-level scoliosis classification is feasible and promising.

Changes in Trunk Appearance After Scoliosis Spinal Surgery and Their Relation to Changes in Spinal Measurements

STUDY DESIGN Retrospective study of surgical outcome. OBJECTIVES To evaluate quantitatively the changes in trunk surface deformities after scoliosis spinal surgery in Lenke 1A adolescent idiopathic scoliosis (AIS) patients and to compare it with changes in spinal measurements. SUMMARY OF BACKGROUND DATA Most studies documenting scoliosis surgical outcome used either radiographs to evaluate changes in the spinal curve or questionnaires to assess patients health-related quality of life. Because improving trunk appearance is a major reason for patients and their parents to seek treatment, this study focuses on postoperative changes in trunk surface deformities. Recently, a novel approach to quantify trunk deformities in a reliable, automatic, and noninvasive way has been proposed. METHODS Forty-nine adolescents with Lenke 1A idiopathic scoliosis treated surgically were included. The back surface rotation and trunk lateral shift were computed on trunk surface acquisitions before and at least 6 months after surgery. We analyzed the effect of age, height, weight, curve severity, and flexibility before surgery, length of follow-up, and the surgical technique. For 25 patients with available three-dimensional (3D) spinal reconstructions, we compared changes in trunk deformities with changes in two-dimensional (2D) and 3D spinal measurements. RESULTS The mean correction rates for the back surface rotation and the trunk lateral shift are 18% and 50%, respectively. Only the surgical technique had a significant effect on the correction rate of the back surface rotation. Direct vertebral derotation and reduction by spine translation provide a better correction of the rib hump (22% and 31% respectively) than the classic rod rotation technique (8%). The reductions of the lumbar Cobb angle and the apical vertebrae transverse rotation explain, respectively, up to 17% and 16% the reduction of the back surface rotation. CONCLUSIONS Current surgical techniques perform well in realigning the trunk; however, the correction of the deformity in the transverse plane proves to be more challenging. More analysis on the positive effect of vertebral derotation on the rib hump correction is needed. LEVEL OF EVIDENCE III.

Automatic detection of microaneurysms and haemorrhages in fundus images using dynamic shape features

This paper presents a novel approach for automatic detection of microaneurysms and haemorrhages in fundus images. First, it begins with a preprocessing stage for shade correction, contrast enhancement and denoising. Second, all regional minima with sufficient contrast are extracted and considered as candidates. Third, in an image flooding scheme, a new set of dynamic shape features is computed as a function of intensity. Finally, a Random Forest classifies the candidates into lesions and non lesions. A set of 143 fundus images with an average of 2210 pixels in diameter was acquired using different cameras and used for training and testing. The proposed approach achieves a global score over the FROC curve of 0.393, while previous work with images of similar resolution reported a score of 0.233.

Assessment of Breast Asymmetry in Adolescent Idiopathic Scoliosis Using an Automated 3D Body Surface Measurement Technique

STUDY DESIGN Cohort study. OBJECTIVES To assess breast asymmetry (BA) directly with 3D surface imaging and to validate it using MRI values from a cohort of 30 patients with significant adolescent idiopathic scoliosis (AIS). Also, to study the influence of posture (prone vs standing) on BA using the automated method on both modalities. BA is a common concern in young female patients with AIS. In a previous study using MRI, we found that the majority of patients with significant AIS experienced BA of up to 21% in addition to their chest wall deformity. MRI is costly and not always readily available. 3D surface topography, which offers fast and reliable breast acquisitions without radiation or distortion of the body surface, is an alternative method in the clinical setting. METHODS Thirty patients with AIS were enrolled in the study on the basis of their thoracic curvature, skeletal and breast maturity, without regard to their perception of their BA. Each patient underwent two imaging studies of their torso: a 3D trunk surface topography and a breast MRI. An automated breast volume measuring method was proposed using a program developed with Matlab programming. RESULTS Strong correlations were obtained when comparing the proposed method to the MRI on the left breast volumes (LBV) (r = 0.747), the right breast volumes (RBV) (r = 0.805) and the BA (r = 0.614). Using the same method on both imaging modalities also yielded strong correlation coefficients on the LBV (r = 0.896), the RBV (r = 0.939) and the BA (r = 0.709). CONCLUSIONS The proposed 3D body surface automated measurement technique is feasible clinically and correlates very well with breast volumes measured using MRI. Additionally, breast volumes remain comparable despite being measured in different body positions (standing and prone) in a young cohort of AIS patients. LEVEL OF EVIDENCE Level IV.

Towards non invasive diagnosis of scoliosis using semi-supervised learning approach

In this paper, a new methodology for the prediction of scoliosis curve types from non invasive acquisitions of the back surface of the trunk is proposed. One hundred and fifty-nine scoliosis patients had their back surface acquired in 3D using an optical digitizer. Each surface is then characterized by 45 local measurements of the back surface rotation. Using a semi-supervised algorithm, the classifier is trained with only 32 labeled and 58 unlabeled data. Tested on 69 new samples, the classifier succeeded in classifying correctly 87.0% of the data. After reducing the number of labeled training samples to 12, the behavior of the resulting classifier tends to be similar to the reference case where the classifier is trained only with the maximum number of available labeled data. Moreover, the addition of unlabeled data guided the classifier towards more generalizable boundaries between the classes. Those results provide a proof of feasibility for using a semi-supervised learning algorithm to train a classifier for the prediction of a scoliosis curve type, when only a few training data are labeled. This constitutes a promising clinical finding since it will allow the diagnosis and the follow-up of scoliotic deformities without exposing the patient to X-ray radiations.

A novel fully automatic measurement of apparent breast volume from trunk surface mesh

This paper presents a novel method for assessing apparent breast volume from trunk surface mesh without any manual intervention. The proposed method requires a closed and smooth triangular mesh of the trunk. It comprises four main steps: automatic nipple localization, automatic breasts delineation, chest-wall interpolation and volume computation. The mean curvature is computed for each vertex using a quadratic fitting approach and used as an indicator to determine the convex fold of the breasts. The delineation is modeled as an ellipse in the frontal plane and all the vertices inside it are removed. The remaining ones are used to interpolate the chest wall with radial basis functions. The voxels inside the resulting mesh without breasts are then subtracted from the original voxelized volume to generate the breasts volume. The validation is conducted on 30 adolescent female for each of which an MRI and a trunk surface (TS) acquisitions were available. Three breast volumes are considered: the anatomical volumes (AV) manually segmented on the MRI, the external volumes computed with the proposed method first in prone position (EVP) using the trunk mesh extracted from the MRI, and second, in standing position (EVS) using the TSs mesh. Significant correlations (R> 0.77) are found between each two of the three volumes. AVs are much larger than both EVS and EPS. In fact, the manual segmentation using MRI slices allows for a direct visualization of the breast posterior delineation. Computed automatically, EVS and EPS are highly similar, indicating that the proposed method is robust to changes from prone to standing position. No significant difference between the regressions on the left and right breasts is noted. Fully-automatic 3D breast volumetry from trunk surface mesh is feasible and provides measurements that are highly correlated to manual MRI volumetry and robust to changes in posture.

Memory efficient and constant time 2D-recursive spatial averaging filter for embedded implementations

Spatial Averaging Filters (SAF) are extensively used in image processing for image smoothing and denoising. Their latest implementations have already achieved constant time computational complexity regardless of kernel size. However, all the existing O(1) algorithms require additional memory for temporary data storage. In order to minimize memory usage in embedded systems, we introduce a new two-dimensional recursive SAF. It uses previous resultant pixel values along both rows and columns to calculate the current one. It can achieve constant time computational complexity without using any additional memory usage. Experimental comparisons with previous SAF implementations shows that the proposed 2D-Recursive SAF does not require any additional memory while offering a computational time similar to the most efficient existing SAF algorithm. These features make it especially suitable for embedded systems with limited memory capacity.

Towards an Automatic Clinical Classification of Age-Related Macular Degeneration

Age-related macular degeneration (AMD) is the leading cause of visual deficiency and irreversible blindness for elderly individuals in Western countries. Its screening relies on human analysis of fundus images which often leads to inter- and intra-expert variability. With the aim of developing an automatic grading system for AMD, this paper focuses on identifying the best features for automatic detection of AMD in fundus images. First, different features based on local binary pattern (LBP), run-length matrix, color or gradient information are computed. Then, a feature selection is applied for dimensionality reduction. Finally, a support vector machine is trained to determine the presence or absence of AMD. Experiments were conducted on a dataset of 140 fundus images. A classification performance with an accuracy of 96 % is achieved on preprocessed images of macula area using LBP features.