Abstract:
The alarming cases of cataract within the last decade
and the projection of cataract cases within the next few decades
call for urgent intervention by early diagnosis. Formal ways of
detecting cataract such as physical examination, tests and
diagnosis are clinic and professional bound. Hence the need for
automation process. Some works have been done on Computer
Aided Diagnosis (CAD) of cataract with tools such as Expert
systems, which are limited to their knowledgebase thus inaccurate.
Early diagnosis of cataract enables quick intervention and
treatment. This paper presents a web-based Computer Aided
Diagnostic for cataract detection system using Convolutional
Neural Network that can be used by any nonprofessional outside
the clinic environment. The systems model trained on a data set
of 100 eye images using transfer learning which were retrieved
from google image search results of “normal human eyes” and
“human eye cataract”. It utilized ImageNet model developed in
ILSVRC2012 using the Convolutional Neural Network classifier
and transferred its knowledge using Transfer learning to train a
new model. The new model gained the ability to classify eye images
into “Normal” and “Cataractious”. The system was designed to
take images as inputs and achieved a Sensitivity of 69%, a
Specificity of 86%, Precision of 86%, F-Score of 56% and AUC of
84.56%. Its accuracy score was 78% which was influenced using
the model trained during the ImageNet image classification using
deep convolutional neural network
