Creating and sharing knowledge for telecommunications

Automatic Cataract Classification based on Ultrasound Techniques using Machine Learning: A comparative Study

Caixinha, M. C. ; Velte, E. V. ; Santos, M. J. S. F. S. ; Perdigão, F. ; Amaro, J. ; Gomes, M. ; Santos, J. B.

Automatic Cataract Classification based on Ultrasound Techniques using Machine Learning: A comparative Study, Proc International Congress on Ultrasonics - ICU, Metz, France, Vol. -, pp. - - -, May, 2015.

Digital Object Identifier: 0

In the present work, ultrasound A-scan signals were acquired from healthy and cataractous porcine lenses. B-mode images were reconstructed from the collected backscattering signals. The parametric Nakagami images were subsequently constructed from the B-mode images. Acoustical and spectral parameters were obtained from the central region of the lens. Image textural parame- ters were extracted from the B-scan and Nakagami im- ages. Ninety-seven parameters were extracted from a to- tal of 75 healthy and 135 cataractous lenses. Lenses with cataract were split in two groups: incipient and advanced cataract, corresponding to a 60 and 120 minutes of im- mersion time in a cataract induction solution, respectively. The obtained parameters were subjected to feature selec- tion with Principal Component Analysis (PCA) and used for classification using four different classifiers: Bayes, K-nearest neighbors (KNN), Fisher Linear Discriminant (FLD) and Support Vector Machine (SVM) classifiers. This paper shows that for the classification of healthy and cataractous lenses (incipient or advanced cataract) the four classifiers show a good performance, with a F- measure ranging from 92.68% to 95.49% for the FLD and KNN classifiers respectively. For incipient versus advanced cataract the SVM classifier shows a higher performance. SVM perform effectively the classification of the cataract severity, with a performance of 90.62% while the perfor- mance of the other classifiers is lower than 80% (72.47% to 79.81% for Bayes and FLD classifiers respectively). Our results showed that SVM can be used as a computer- aided diagnosis (CAD) system for the cataract classifica- tion based on the ultrasound analysis.