Microaneurysms detection in color fundus images using machine learning based on directional local contrast

Abstract

Background: As one of the major complications of diabetes, diabetic retinopathy (DR) is a leading cause of visual impairment and blindness due to delayed diagnosis and intervention. Microaneurysms appear as the earliest symptom of DR. Accurate and reliable detection of microaneurysms in color fundus images has great importance for DR screening.

Methods: A microaneurysms' detection method using machine learning based on directional local contrast (DLC) is proposed for the early diagnosis of DR. First, blood vessels were enhanced and segmented using improved enhancement function based on analyzing eigenvalues of Hessian matrix. Next, with blood vessels excluded, microaneurysm candidate regions were obtained using shape characteristics and connected components analysis. After image segmented to patches, the features of each microaneurysm candidate patch were extracted, and each candidate patch was classified into microaneurysm or non-microaneurysm. The main contributions of our study are (1) making use of directional local contrast in microaneurysms' detection for the first time, which does make sense for better microaneurysms' classification. (2) Applying three different machine learning techniques for classification and comparing their performance for microaneurysms' detection. The proposed algorithm was trained and tested on e-ophtha MA database, and further tested on another independent DIARETDB1 database. Results of microaneurysms' detection on the two databases were evaluated on lesion level and compared with existing algorithms.

Results: The proposed method has achieved better performance compared with existing algorithms on accuracy and computation time. On e-ophtha MA and DIARETDB1 databases, the area under curve (AUC) of receiver operating characteristic (ROC) curve was 0.87 and 0.86, respectively. The free-response ROC (FROC) score on the two databases was 0.374 and 0.210, respectively. The computation time per image with resolution of 2544×1969, 1400×960 and 1500×1152 is 29 s, 3 s and 2.6 s, respectively.

Conclusions: The proposed method using machine learning based on directional local contrast of image patches can effectively detect microaneurysms in color fundus images and provide an effective scientific basis for early clinical DR diagnosis.

Keywords: Color fundus image; Directional local contrast; Feature extraction; Machine learning; Microaneurysms’ detection; Patch.

Fig. 1

Color fundus image of DR

Fig. 2

Illustration of the proposed MA detection method

Fig. 3

ROC curves of three classifiers on the two databases. a e-ophtha MA database. b DIARETDB1 database

Fig. 4

ROC curves of three classifiers on the two databases without using DLC feature. a e-ophtha MA database. b DIARETDB1 database

Fig. 5

FROC curves of different methods for MA detection on the two databases. a e-ophtha MA database. b DIARETDB1 database

Fig. 6

Lesion level evaluation for MA detection results on e-ophtha MA database. a Results of MA detection, where green circles indicate TPs, white circles indicate FPs, and red circles indicate FNs; b examples of TP and FN; c examples of FP

Fig. 7

Lesion level evaluation for MA detection results on DIARETDB1 database. a Results of MA detection, where green circles indicate TPs, white circles indicate FPs, and the red circle indicates FN; b examples of TP and FP; c examples of FP and FN

Fig. 8

Analysis of MA detection results compared with ground truth on DIARETDB1 database. a MA detection results corresponding to different labeling confidences, where yellow circles indicate labels with confidence $\ge 75\%$, orange circles indicate labels with confidence $\ge 50\%$, and brown circles indicate labels with confidence $\ge 25\%$. b Evaluation of MA detection results with ground truth of confidence $\ge 75\%$, where green circles indicate TPs, white circles indicate FPs, and the red circle indicates FN, corresponding to Fig. 7

Fig. 9

Process of blood vessels segmentation. a Green channel image; b result of shade correction ($I_{\text{sc}}$); c BV enhanced image (green circles indicate MAs); d preliminary BV segmentation (green circles indicate MAs); e final result of BV segmentation

Fig. 10

Process of MA candidate regions extraction. a Result of shade correction ($I_{\text{sc}}$); b result of preprocessing ($I_{\text{gauss}}$); c result of BV removal ($I_{\text{gsBV0}}$); d result of contrast stretch ($I_{\text{CS}}$); e preliminary MA candidate regions ($I_{\text{can1}}$); f final result of MA candidate regions ($I_{\text{can}}$, green circles indicate ground truth of MAs)

Fig. 11

Examples of patches. a MA patches; b non-MA patches

Fig. 12

Comparison of different structures in 25×25 patch. a MA; b HM; c BV; d background

Fig. 13

DLC distribution on different structures shown in Fig. 12, where radius indicates the DLC value along the direction angle $\theta$