advertisement

Editors Selection IGR 7-1

Structure and function measurements: Machine learning classifiers

Comment by Christian Mardin on:

20792 Bayesian machine learning classifiers for combining structural and functional measurements to classify healthy and glaucomatous eyes, Bowd C; Hao J; Tavares IM et al., Investigative Ophthalmology and Visual Science, 2008; 49: 945-953

Find related abstracts

Bowd et al. (457) investigated Bayesian machine learning classifiers for combining structural and functional measurements to classify healthy and glaucomatous eyes the ability of machine learning classifiers to improve glaucoma diagnosis by combining structural and functional measurements. The number of patients is well chosen for this paper. It would be interesting to compare the results of the learning group with an independent test group. The definition of early glaucoma with 75% of patients with a md > -4.08 in SAP is discussable as we know, that more than 30% of rim area is lost at that stage. Nevertheless the authors combine OCT for the measurement of RNFL and SAP, which in practice is extremely important as both methods are standard. OCT will in future play a more important role in early diagnosis, as the resolution is improving (e.g., spectral-domain-OCT).

The combination of both methods reached higher values in ROC in glaucoma diagnosis than single measurements (OCT, SAP) as none of the methods alone identifies glaucoma in every patient. This has been shown in the past also with GDx and SAP also. The results are well presented with all the necessary information. The relatively small difference between the single OCT and SAP method to the combination may be the result of the fact, that OCT is rather sensitive in diagnosing perimetric glaucoma and as visual field was a definition criterion for glaucoma. It must also be kept in mind, that between structural and functional defects there is in some patients a time gap and therefore the dependency between both is not always visible in a cross-sectional study. Another point which has to be kept in mind is, that there is an age-RNFL correlation and in this study the controls were significantly younger than the glaucomas. This may lead to overoptimistic results.

• ← Previous
• Next →