Automated drusen detection in retinal images using analytical modelling algorithms

Abstract

BackgroundDrusen are common features in the ageing macula associated with exudative Age-Related Macular Degeneration (ARMD). They are visible in retinal images and their quantitative analysis is important in the follow up of the ARMD. However, their evaluation is fastidious and difficult to reproduce when performed manually.MethodsThis article proposes a methodology for Automatic Drusen Deposits Detection and quantification in Retinal Images (AD3RI) by using digital image processing techniques. It includes an image pre-processing method to correct the uneven illumination and to normalize the intensity contrast with smoothing splines. The drusen detection uses a gradient based segmentation algorithm that isolates drusen and provides basic drusen characterization to the modelling stage. The detected drusen are then fitted by Modified Gaussian functions, producing a model of the image that is used to evaluate the affected area.Twenty two images were graded by eight experts, with the aid of a custom made software and compared with AD3RI. This comparison was based both on the total area and on the pixel-to-pixel analysis. The coefficient of variation, the intraclass correlation coefficient, the sensitivity, the specificity and the kappa coefficient were calculated.ResultsThe ground truth used in this study was the experts' average grading. In order to evaluate the proposed methodology three indicators were defined: AD3RI compared to the ground truth (A2G); each expert compared to the other experts (E2E) and a standard Global Threshold method compared to the ground truth (T2G).The results obtained for the three indicators, A2G, E2E and T2G, were: coefficient of variation 28.8 %, 22.5 % and 41.1 %, intraclass correlation coefficient 0.92, 0.88 and 0.67, sensitivity 0.68, 0.67 and 0.74, specificity 0.96, 0.97 and 0.94, and kappa coefficient 0.58, 0.60 and 0.49, respectively.ConclusionsThe gradings produced by AD3RI obtained an agreement with the ground truth similar to the experts (with a higher reproducibility) and significantly better than the Threshold Method. Despite the higher sensitivity of the Threshold method, explained by its over segmentation bias, it has lower specificity and lower kappa coefficient. Therefore, it can be concluded that AD3RI accurately quantifies drusen, using a reproducible method with benefits for ARMD evaluation and follow-up.

Adaptive Super-Candidate Based Approach for Detection and Classification of Drusen on Retinal Fundus Images

“…5(b). The optimal performance points occur at [5] drusen detection -, 23 Sens, Spec -> 96% except 1 image (88%) Brandon et al [2] drusen detection STARE,113 Sens -86%, Spec -93%, Acc -87% discrimination (size) STARE, 113 Acc -71% Mora et al [6] drusen detection -, 22 Sens -74%, Spec -94% discrimination (area) -, 22 Sens -63%, Spec -96% Smith et al [1] discrimination (area) -, 358 Sens -87.3%, Spec -95.3% Wong et al [11] drusen detection THALIA,350 Precision -95.46 ± 0.94 Kose et al [7] discrimination ( [12] drusen detection CAPT Sens -82%, Spec -75%, Acc -80% Barriga et al [8] HD/SD discrimination -, 5 Acc -96% for 120 patches (5 images) Proposed method drusen detection STARE, ARIA, 214 Sens -80%, Spec -90%; Sens -95%, Spec -70%,AUC -0.9257 HD/SD discrimination STARE, ARIA, 214 Sens -80%, Spec -92%; Sens -95%, Spec -70%, AUC -0.915 OP DET1 (sens−80%, spec−90%), OP DET2 (sens−95%, spec−70%) for drusen detection at the confidence score threshold value of 0.4, 0.25 and OP DISC1 (sens − 80%, spec − 92%), OP DISC1 (sens − 95%, spec − 70%) for HD and SD discrim- Table 1, our method gives good performance for drusen detection and discrimination.…”

“…Moreover, automatic drusen detection is hindered by normal anatomical structures such as retinal nerve fiber layer (RNFL), optic disc and choroidal vessels in tessellated retina and adverse imaging conditions such as non-uniform illumination and low contrast. Therefore various preprocessing techniques such as iterative background modeling [1], [6], multi-level histogram equalization [5] has been developed in prior work. While identification of hard drusen is well attempted and utilized directly in the AMD grading systems, reported literature on the detection of soft drusen are scarce.…”

“…Utilizing this difference in size, classification of drusen has been done in prior work based on size [7], [2] and area [1], [6]. However, reliability of size and area measurement is limited by factors such as angle of acquisition (magnification), inaccurate segmentation and clumped drusen being misclassified as isolated drusen.…”

A Type-2 Fuzzy Model Based on Three Dimensional Membership Functions for Smart Thresholding in Control Systems

“…We selected Cubic Smoothing Spline (CSS) for generate the upper and lower membership functions because of non-uniform illumination of the Three Dimensional Memebership Functions (3DMFs). In the Type-2 domain, the estimation of the 3DMFu and 3DMFL are exanimate from the fitting of a cubic smoothing Spline, ( Mora et al,2011) to the 3DMF(x,y). The select CSS is a special class of Spline that can capture the low 3DMF value that limited the non-uniformity of the 3DMF (Culpin, 1986).…”

“…Similar 3DMF presented in section 4.1, to generation of polar MF, CSS is used. The estimation of the MF also exanimate from the fitting of a cubic smoothing Spline, (Mora et al,2011) to the 3DPMF(r, ). The fitting objective is to minimize the equation.…”