Digital photography has become a tool frequently used in dermatology and cosmetology. Initially limited to illustration purposes, this unexpensive technique has gradually found its usefulness in characterization and quantification. Many investigations in extremely different situations are now made possible with digital photography associated with computing and image analysis, compared with reference measurements, and completed by clinical scoring (follow-up of acne, assessment of wrinkles, psoriasis, pigmented marks, follow-up of wound healing...).
However, digital photography remains a technique for specialised trained technicians. Image analysis is particularly complex, insufficiently documented, hence hardly reproducible. An investigation carried out with digital assessment only is often considered as unreliable because its flexibility and extent can lead to excesses and deception. With data compression techniques and digital retouching, it is almost impossible to detect the exact origin of an image or guarantee its originality.
To give a function to an image other than its conventional use as an illustration, we try to extract parameters that can be quantified to characterize an object. To obtain these results, two conditions are required : to have a clearly or partially identified signal in the image (a sufficient ratio signal/noise), and to define an image analysis process, i.e. an algorithm to extract and summarize this information. The first prerequisite is not obvious because the investigators tend to disagree about the diffuse, informal and subjective object they want to quantify.
Image analysis can only quantify what the investigator can see and what generates adhesion. It cannot bring a solution to an imperfectly defined problem. The respect of these constraints allows to formulate an efficient algorithm leading to the calculation of intuitive and rational parameters. The analysis will then be confirmed by confronting it to medical expertise. Particular attention must be paid to the choice of parameters which have to be defined and documented for appropriate evaluation and improvement.
Photography in crossed polarized light enhances the saturation of colours, thus reducing the reflection of the lighting. Its application to the study of acne is interesting in this particular example.
In digital photography, quantification brings up three problems : calibration, standardisation and repositioning.
To calibrate an image is to give it a scale, a physical correspondence between the pixels that form it and any physical entity. Calibration is therefore an essential preliminary step before any analysis. It is usually achieved with a colour chart or graded scale positioned near the photographed object. Such a procedure lacks precision though (the object and the camera must be stationary, and the depth of field strictly identical) especially with regard to the resolution of recent devices (CCD sensor of 6.1 to over 10 million pixels, providing images up to 3000x3000 pixels). Experience shows that it is illusory to try and measure differences below 1 mm (with conventional objectives) and the variation rate is usually between 10 and 15 %.
Another difficulty is to define and fix the parameters of image taking definitely. Standardisation implies the control of all the conditions in which the photograph is taken in order to ensure that the changes detected in the image are not induced by causes intrinsec to the study. The position and posture of the subject, the room, the lighting, the distance and the type of camera are parameters which can vary and must therefore be defined precisely. The most recent devices hold digital image processors with very elaborated algorithms which take into account the analysis of light and exposure. The comparison of photographs from different types of cameras is difficult because of the adjustments selected by the users as well as by the manufacturers. Lighting is of prime importance in research on colour and texture. A flashlight can enhance or inversely reduce the contrast and colour balance of a photograph and make it inexploitable. For this reason, recent studies tend to use filtering techniques and artificial light. The sources of variation are so numerous that they are unacceptable in multicentric studies.
Finally, an additional issue is repositioning : it is the possibility to find very accurately the zone or the orientation of the object photographed as a reference. In most studies, repositioning is made several weeks or months after T0. Variations are lessened by using supports and stereostatic tables which maintain the subject stationary, associated with a remote control of the camera. Computer solutions are available to compensate these alterations in space and require specific processes (repositioning, autocorrelation...). This issue is fundamental in clinical trials because it concerns most of the digital imaging devices.
Geometrical repositioning includes 3 steps : 1) manual selection of pairs of points 2) connection of these points together and creation of a geometrical transformation 3) application of the transformation to the image
These steps allow to correct bidimensional translation (horizontal or vertical variations of the volunteer's position), rotation or homothetia (distance variations between the volunteer and the sensor).
The investigator faces other problems : the visualisation of the images depends on the monitors calibration. The assessement of colour saturation presents problems which vary according to the type of evaluation : from a person, a monitor or a photograph. Acually each stage of the whole digital process is likely to present difficulties (data capture, visualisation and impression).
Digital photography is a promising tool and its mastership and use will probably progress in the near future. However performances and technical advances must not hide the intellectual strictness necessary to exploit this technology. Image analysis must evolve from its present place in the computing specialist sphere and become exposed to review. This stage involves better training, documentation about algorithms and the parameters used in image analysis, and a real willingness for discussion and exchanges.