A data mining framework for the classification of retinopathy images based on a new multistage prediction algorithm

Karim Hashim AL-Saedi, Herbert F. Jelinek

Research output: Contribution to journalArticlepeer-review

30 Downloads (Pure)


Medical image processing, analysis and classification is a rapidly expanding field providing possible solutions for health care providers including ophthalmologists and optometrists. To be of value, image analysis and classification requires high accuracy and fast processing. Early detection of diabetic retinopathy can lead to better treatment outcomes, especially in rural and remote areas where there is a lack of specialists. In the current work we propose a highly accurate prediction model based on optic disc color characteristics. We propose a data mining algorithm based on a top-down processing framework. The framework involves a new Multistage Prediction algorithm (MSP1) consisting of segmentation of the optic disc, dilation, and color normalization, color histogram determination, and calculating the predicted classification score of each image. The final step carries out the process of classification of all images based on the Group Method of Data Handling(GMDH) application. One hundred and fifty seven images were available for classification. The results indicate that the proposed Multistage Prediction algorithm combined with the GMDH classification framework improved on previous results with an overall accuracy of 96.8%, and sensitivity of 95% and an F-measure for the classification performance of96%. MSP1 is easy to implement on any laptop and therefore provides a robust option for tele-ophthalmology diagnostics of retinopathy.
Original languageEnglish
Pages (from-to)4201-4206
Number of pages6
JournalInternational Journal of Engineering and Technology
Issue number4
Publication statusPublished - Dec 2018


Dive into the research topics of 'A data mining framework for the classification of retinopathy images based on a new multistage prediction algorithm'. Together they form a unique fingerprint.

Cite this