Automated detection of microaneurysms using scale-adapted blob analysis and semi-supervised learning

Abstract

Despite several attempts, automated detection of microaneurysm (MA) from digital fundus images still remains to be an open issue. This is due to the subtle nature of MAs against the surrounding tissues. In this paper, the microaneurysm detection problem is modeled as finding interest regions or blobs from an image and an automatic local-scale selection technique is presented. Several scale-adapted region descriptors are then introduced to characterize these blob regions. A semi-supervised based learning approach, which requires few manually annotated learning examples, is also proposed to train a classifier to detect true MAs. The developed system is built using only few manually labeled and a large number of unlabeled retinal color fundus images. The performance of the overall system is evaluated on Retinopathy Online Challenge (ROC) competition database. A competition performance measure (CPM) of 0.364 shows the competitiveness of the proposed system against state-of-the art techniques as well as the applicability of the proposed features to analyze fundus images.

Incorporating Spatial Information for Microaneurysm Detection in Retinal Images

“…The obvious advantage of an unsupervised method is that they do not require a training phase. Some initial candidate detection methods that have been proposed are Gaussian filters [4]- [6] or their variants [8], [17], [18], simple thresholding [15], [16], [19], Moat operator [20], double ring filter [9], mixture model-based clustering [21] 1D scan lines [13], [14], extended minima transform [11], [22], Hessian matrix Eigenvalues [7], [23], Frangi-based filters [24] and hit-or-miss transform [10].…”

Automatic Detection of Microaneurysms for Diabetic Retinopathy Screening Using Fuzzy Image Processing

“…The Retinopathy Online Challenge (ROC) presents an online competition for numerous methods in microaneurysms detection to compare with each other on the same data [8]. Other recent works on the detection of microaneurysms are [9][10][11][12].…”

Microaneurysm detection in retinal images using an ensemble classifier

“…A large portion of the methods rely on a Gaussian matched filter [14]- [16], or a variant of the Gaussian filter [7], [17], [18] in order to detect the initial set of candidates. Other methods for initial candidate detection include thresholding [8], [9], [19], Moat operator [20], double ring filter [21], mixture modelbased clustering [10] 1D scan lines [4], [5], extended minima transform [22], [23], Hessian matrix Eigenvalues [24], [25], Frangibased filters [26] and hit-or-miss transform [27]. A variety of classification techniques have been used in order to reduce the number of false positive detections.…”

“…A variety of classification techniques have been used in order to reduce the number of false positive detections. These include Linear Descriminant Analysis (LDA) [14] K-Nearest Neighbours (KNN) [15], [16], [18], [25], Artificial Neural Networks [21], [27], Naive Bayes [22] and Logistic Regression [28]. A number of techniques did not rely on a classifier (unsupervised methods) [4], [5], [8], [9] .…”

“…However we found that this feature is not discriminative enough and does not detect many candidates that lie on vessels. Therefore we introduced two other features: isvessel_loose (25) and isvessel_QUARTZ (26). isvessel_loose is an adaptation of Fleming's isVessel feature with more relaxed constraints.…”