Page 177 - Computational Retinal Image Analysis

172 CHAPTER 10 Statistics in ophthalmology

as intraocular pressure, nowadays we have the opportunity to go so much further
with much richer and more complex data sets such as retinal images. The explo-
sion of data (which perhaps exceeds the limits of traditional statistical methods)
has led to increasing analyses being conducted by those without formal training
in statistical methodology. Good statistical principles still apply, however, and for
these reasons for a modern researcher in ophthalmology a working knowledge of
traditional and more modern statistical methods is indispensable. Statistical er-
rors in medical research can harm the very people that researchers seek to help.
Computational analysis has this same potential, which is why the learnings from
traditional statistical methodology must be handed over.
Our primary goal in this chapter is to give a comprehensive review of the current
most important design paradigms and data analysis methods used in ophthalmol-
ogy and show briefly how all the concepts link together within the discipline of
the statistical science. We have a second goal. This comes from our experience of
reading numerous published reports that are analyzed by people who lack training
in statistics. Many researchers have excellent quantitative skills and intuitions, and
most times they apply statistics correctly. Yet, there are situations when statistics
or design is not done properly, or the data are not used to the full potential which
may be due to a simple lack of statistical training. Some statistical concepts are
challenging and if introduced too early or without warning that the methods they
use might take time to sink in leave people hating the very word statistics. This is
a huge shame. Clinical researchers have the clinical expertise to translate findings
from complex statistics to have meaning. Computer scientists have the skills to pro-
gram methods that are more complex than traditional statistical techniques needed.
Different disciplines need to engage with each other in order to best use the innova-
tive data explosion that we face. Different disciplines develop their own terminol-
ogy but it is not new science if in reality the same discovery is simply being made
within a different academic discipline with a new name. It is a wasted opportunity.
If instead of re-inventing the wheel we hand over learning by speaking with each
other, we will go further with our discovery. With increasing use of data science and
machine learning (ML) methods to analyze imaging data and to illustrate data, we
aim to try to establish a link between statistical science and retinal image analysis
and machine learning: for instance, what are the differences and commonalities
between statistics and ML? Is it worth discussing ML techniques for retinal data
analysis, from the point of view of statisticians?

1.2 The contribution of statistics in ophthalmic and vision research

Some researchers believe that statistics is a collection of methods for design of
studies and for analysis of data such as t-test, analysis of variance, linear regres-
sion etc. This is true in part. Statistics is also a scientific discipline where statisti-
cians strive for something deeper, for the development of statistical methods of
study designs, data collection, data analysis and communication of results. They