Interactive dynamic graphical techniques for the exploration of functional data

The use of interactive dynamic graphics has become a common practice for the exploration of multidimensional data sets. The availability of powerful and inexpensive hardware and software for graphical computing makes the use of such techniques feasible for the examination of complex forms of data. This paper describes some simple techniques, which were implemented in the LISP-STAT environment, for the visualization of functional data arising from studies of optical technologies used for the detection of cervical intraepithelial neoplasia or squamous intraepithelial lesions.

The methods demonstrated have been implemented in software coded in LISP-STAT, a free statistical computing package available for most computer systems. The data used in this paper are drawn from a previous study in which fluorescence spectroscopy was measured from cervical sites at 337 nm, 380 nm, and 460 nm excitation in cervical screening patients. The goal of the project is to explore biographical variables to better understand the biology of fluorescence.

199 measurements were taken in 55 women with normal Pap smears. The data are recorded as spectra showing the intensity of emission excitation versus emission in nanometers. Covariate variables available for analysis are current smoker vs. nonsmoker premenopausal vs. postmenopausal, tissue type (columnar, squamous, and transition zone), and age in years. Although the optical measurements show consistent changes between normal and abnormal tissue in individual patients, there is wide variation in the intensity of the measurements between patients, even for normal tissue. Patient age affects the fluorescent spectrum showing increasing intensity with increasing age. Menopausal status affects the fluorescent spectra coincidentally with age. Smoking and race do not appear to affect the spectra in this sample of patients.

The use of interactive graphical techniques permits the data analyst to examine multidimensional data in intuitive ways. These explorations allow non-statisticians to explore the data in a perceptive manner that may lead to new approaches in algorithm development for optical technologies.