Sumeet Dua and Mohit Jain

3.4.2.1. Using classifiers

In order to choose to use a classifier, a clinician must first construct a model. In model construction, clinicians build or train a model according to the training data; in model usage, clinicians classify future or unknown objects or estimate accuracy. In ophthalmology, clinicians classify how many images have been correctly classified.

3.4.2.2. K nearest neighbors

KNNs are used to find the nearest neighbors of the input image or those images that have the highest probability of being matched with the input image. K can be any integer value. To find the nearest matches, clinicians can use similarity measures such as Euclidean distance or dynamic time warping, which gives the distance between the two images. If the distance between the two images is zero, then they are matched exactly.

3.4.2.3. GUI of the system

The main menu of the GUI will include options, for example image acquisition, feature extraction, and classification, which the clinician can choose. This automatic and interactive system aids healthcare professionals in using the decision support system and gives them important information to avoid the risk of a patient developing a visual disability.

3.4.3. Computational Decision Support System: Glaucoma

Glaucoma is an eye disease that affects the optic nerve, and, if untreated, can lead to blindness. There are two categories of glaucoma: open-angle glaucoma and closed-angle glaucoma. Closed-angle glaucoma is painful and can lead to blindness quickly. Open-angle glaucoma progresses slowly, and has often developed significantly, before the patient learns he or she has it.

Open-angle glaucoma is generally caused by increased pressure within the eye, and it is difficult to find symptoms at the early stage. Thus, regular checks are necessary to avoid this disease. Damage to the optic nerve, a side effect of glaucoma, cannot be reversed; thus, early treatment is important to the maintenance of good eye health. Early detection can be

108

Computational Decision Support Systems and Diagnostic Tools

maintained by reducing intraocular pressure (IOP). Once the disease is detected, medications can help to stop the progress of the diseases. Therefore, it is important to diagnose glaucoma early to minimize the risk of blindness. Factors to consider when performing glaucoma screening are sufficient sensitivity and specificity, cost effectiveness, the quickness of the diagnosis, and high quality equipment.

Sensitivity and specificity are statistical measures of the performance of a test. Sensitivity measures whether a person with glaucoma is correctly identified. Specificity measures whether a person with healthy eyes is correctly identified. Glaucoma screening should be economical to avoid stress or patient reluctance to do a screening. Diagnosis should be efficient and effective, as glaucoma treatment and medication is expensive. Equipment should be easy to use and should give precise results. Therefore, an automatic and interactive computational system that can help avoid visual loss caused by glaucoma is necessary.

Fuzzy logic and classifiers are computational system methodologies for efficiently diagnosing glaucoma.28−30

3.4.3.1. Using fuzzy logic

The Heidelberg retina tomograph (HRT), which allows 3D images of the retina and the topography of the optical nerve head to be obtained, can aid in the acquisition of images that can be analyzed over the time. Once images are acquired, fuzzy logic can be applied in three steps: image preprocessing, visual field examination, and intra ocular pressure (IOP).

First, in glaucoma screening, appropriate image processing is applied to enhance the visibility of retinal nerve fiber layer defect (RNFLD) features that occur during early-stage glaucoma. The image processing shows the images of the optic disc. Image enhancement methods include loading the images, converting the RGB images into gray-scale, resizing the images, applying an image histogram so that all the images have uniform brightness or intensity, and removing the noise from the images using filtering techniques, such as Gaussian or adaptive filtering.

Second, a visual field examination is performed. In a visual field examination, different areas of different colors and shapes are scattered over the visual field. The technician uses the computational unit to make an area flicker, and the patient is asked to close one eye and press a key on the

109