TELEPATHOLOGY

O.FERRER-ROCA
Catedra Anatomia Patologica. Facultad Medicina. Tenerife. Canary Islands. Spain.

INTRODUCTION

The pathologist based their diagnosis on colour-stained specimens and identification of "controlled artefacts" induced in an otherwise transparent and not visually detected specimens.

FIGURE 3.1 . Shows the logarithmic correlation between Specific Videconsultation experience (VC) and diagnosis efficiency based on the results of the ROC area evaluation. Ferrer-Roca et al [6]

The training period required to modify this previous control scenario to a more constrained one (see also Annex IX) in which sampling, focus and colours are much more limited (such in Teleconsultation cases) is around 1,5 years [6], and have to be taken into account when the so called "expertness consultation" is considered.

APPLICATIONS

Telepathology applications can be divided according the management and interactivity of images into:

A) STATIC TELEPATHOLOGY (Teleconsultation)
Done with Static Images sent for consultation in various ways (ftp, www, static images under videotelephony, modem, etc..).

B) KINETIC TELEPATHOLOGY
That includes the capability to monitor the microscope at distance in order to do the sampling. Images are sent in full resolution either as static images, as live highly compressed images (i.e. through Videoconferencing standards) or both at the same times.(i.e for Intraoperative specimens).

C) DYNAMIC TELEPATHOLOGY
That includes plus the B) option, the capability of having full colour non-compressed images in real time (Live).

REQUIREMENTS

Table 3.4. Telepathology requirements

1.- MULTIMEDIA DATA BASE

Is a preliminary condition, since no diagnosis can be done without a patient`s clinical history and knowledge of previous biopsies. When previous biopsies are archived in the referral laboratory, they should be available (in the data base) for comparison purposes. An efficient method of query or selection per patient and type of diagnosis is important.

2.- COLOUR IMAGES OF SUFFICIENT RESOLUTION

2.1.- DYNAMIC RANGE:

The dynamic range of a system is defined as the difference between the saturation level and minimum detection level divided by the capability of detection change.

Diagnosis in pathology is based on colour images, that with 8 bits of dynamic range may produce sufficient information; the reason being that pathology acquisition systems are based in Videocameras that have an implicit non-linear response (gamma correction). This effect on the image is inverted by the gamma correction of the display systems obtaining a final effect of an adequate visual perception -see further details in Annex IV and Chapter 2 (Displays).

2.2.- SPATIAL RESOLUTION

In the digital sampling of pathology images the spatial resolution is of paramount importance in the accurate visual perception, and therefore:

a) Introduces variability on the minimum requirements for a capture system depending on the microscopic power used.
b) Is directly related to the capability of pathology images to support lossy compression algorithms.

SAMPLING THEORY

Spatial frequency is related with the amount of information present in a one spatial dimension unit. Spatial resolution is defined as the maximum spatial frequency (Fmax) able to be detected or transmitted; for digital images, it represents the two dimensional pixel matrix of the acquisition device.

Sampling theory tries to reduce to a reasonable limit the amount of information that should be stored and processed to avoid heavy computation processes and optimise transmission times over the network. According to the Sampling Theory of Shannon (sample frequency or distance), in order to assure that a discrete sample (digital) will unequivocally reproduce an analogue (continuos) image, samples should be taken at 2Fmax. It means double of the maximum spatial frequency; all this limited by the signal to noise ratio of the system.

That is the reason why 512 x 512 image sampling at higher power can provide good quality images (some investigators mention 1Kx1K requirements (Black-Schaffer et al. 1995)) but that could be unacceptable for lower power sampling, where the amount of information per space unit is much higher.

The high resolution cameras present in the market (still too expensive) directly record digital images with very high spatial resolution. Two main consequences are derived from this technique:

• A recorded image at low microscopic power can support digital zooming simulating higher power microscopic views.
• The image is not affected by the gamma correction of the Video-systems (Annex IV) -depending on the cameras- and is therefore comparable to a digital radiography from the visual perception and image processing point of view.

2.3. COMPRESSION METHODS

Address the issue of how to reduce the amount of data without a sensible loss of important information. The term "important", varies according to the subsequent analysis procedure: a) Visual inspection or diagnosis; b) Image analysis; c) Image quantization

Include techniques of luminance and colour reduction (reduction of dynamic range), spatial resolution reduction, or both, as well as simple data reduction.

• A.- Colour reduction techniques:

A.1. Through YUV encoding. Usually based on colour sampling reduction of 8,4,4 bits information acquisition.

A.2. Reduction of colour palette to 256 colours (8 bits).

A.3. Median cut colour quantization. It is a colour quantization technique that optimises the representation of the original colour in the final palette.
This technique produces median cuts of colours on RGB to arrive to 256 colours (8 bits), therefore, the most frequent colours have greater range of colour gradation than infrequent colours. This can be further optimised by dithering that appears to expand the available colour palette by juxtaposing pixels of different colours to create the illusion of additional colours by visual blending. In diffusion dithering this is accomplished by calculating the numeric difference between the original and final colour of each pixel and distributing this difference among the neighbouring pixels.

• B.- Spatial resolution reduction: Includes sampling reduction with or without inter-pixel interpolation e.g. digitise one every two pixels.
• C.- Data reduction as the ones included in the classical lossy and lossless compression methods.

The temptative DICOM standard for pathology images already described in Chapter 2, that supports JPEG lossy and loosless compression, is under study.

3.- INTERACTIVE CONTROL OF COLOUR

Misinterpretation of colour at the receiving point is an important cause of error together with sampling problems. Therefore, a minimum knowledge of the Colour Theory is required to understand the need for interactivity due to the various Spectral Responses as well as built-in Gamma Correction that cameras and displays provide, and that affects the colours and visual perception. See Colour Theory in Annex IV.

4.- CONTROLLED SAMPLING

Diagnostic discrepancies due to sampling error are in the range of 9 to 6,3%. The sampling control can be done by different procedures. For microscopic views the more advanced one is the robotics microscopic systems (neither expensive or technically complex since many microscopes have already a motorised stage) in which sampling can be done at the distant site; the simplest way encompasses sending low power images with marks where the higher power images will be taken. The same statement can be applied to macroscopically images using a motorised surveillance camera

The sampling possibilities are as follows:

4.1. Sampling done by another person.
4.2. Sampling self-taken (Robotics Microscope).
As a minimum requirement we have to include an application that allow us to assure where are we located in the specimen, as well as which parts of the specimen are already analysed.
4.3. Virtual slide or complete specimen digitisation.
a) Take all fields at a higher power (the minimum requirement of point 5.2 is needed).
b) Low power capture with a high resolution camera that will allow thereafter a digital zooming.

Liability formulas: In any case diagnosis based at distance that do not allows robotics or interactive sampling should include the statement:
"based on the images received"

and in many of the real overseas consultations also include the statement:
"if confirmed and contra-signed by the responsible pathologist"

5.- SECURITY AND CONFIDENTIALITY TOOLS

It is mandatory: a digital signature, the authentication of data and control and the register of any modification in a diagnostic system which is based on the data available (see Chapter 8).

This will join with the Organisational Requirements to use the system by a staff that have to be aware of the information protection techniques for manipulating personal sensible data, as well as liability issues linked to the task (see Chapter 9). As part of the organisational requirements is the Appointment Management that although not strictly required, is important for an efficient and functional "on line" supervision or diagnostic service. This can be done:

- through a STUDIO organisation (see Chapter 7) or
- on a personal computer based system, by electronic means.

Finally, the Computer Supported Cooperative work -CSCW, does not seem to be an essential feature for a Telepathology diagnostic system. Nevertheless, this will be the ideal situation for an everyday work supervision, comparable to the multiple-head microscope discussions between senior and junior pathologists.

TELEQUANTIZATION AT DISTANCE

Especial mention requires the application of Pathology that implements distant quantization such as: Morphometry, Immunohistochemistry of oestrogen receptor, DNA quantization, etc.

Table 3.5. Requirements in Telequantization

1.- Image reception quality assures:

Gain and offset control (for densitometry).
Gamma correction control (for colour evaluation).
Geometric control location and resolution (for segmentation and morphometry).
See further details in Annex IV.

2.- Robust segmentation algorithms:

Based on:

Screen segmentation. Individual object segmentation.

3.- Robust and quick algorithms:

Of Image processing and quantization.
Require also quick algorithms adapted to the image quality.

It is obvious that to assure an equal distant quantization it is required a minimum knowledge of basic image analysis as well as how the connected devices operate, which is sometimes a problem difficult to solve. For example:

• in order to assure a reduction of the noise, if one of the devices does an integration or adding up image algorithm (low pass filter effect) the number of adding images have to be controlled in order not to loose information.
• it has to be controlled that the background is treated in the same way.
• gamma correction and image format handling have to be controlled in both sides -See Annex IV-.

Detailed scientific information of distant quantization is very limited in the literature [8,9,10].

References

1. Black-Schaffer S., Flotte T.J.: Current issues in Telepathology. Telemedicine J. 1:95-106,1995
2. Doolittle MH., Doolittle KW., Winkelman Z. and Weinberg DS. Colour Images in Telepathology: How many colours do we need ?. Human Pathology 28: 36-41,1997
3. Weinstein RS., Bloom KJ., Roze S. Telepathology: Long-distance diagnosis. Am.J.Clin.Pathol. 91: S39-S42,1989 (supp 1)
4. Prossie J. Colour optimisation and dithering. PC Magazine March 28:253-257,1995.
5. Gonzalez R.C., Woods RE. Digital Image Processing. Addison-Wesley Publ Comp Inc 1992. Reading Mass.USA:
6. Ferrer-Roca O., Pitti S., Diaz-Cardama A., Markidou S., Maeso MC, Ramos A., Coen H. Factors influencing distant tele-evaluation in Cytology, pathology, conventional radiology and mammography. Anal.Cell Pathol 10:13-23,1996
7. Raab SS., Robinson RA, Snider TE et al. Telepathology review: Utility diagnostic accuracy and interobserver variability on a difficult case consultation service. Mod Pathol. 10: 630-635, 1997
8. Ferrer-Roca O. Videotelefono en Anatomia Patologica. Medicion de DNA y Receptores Inmunohistoquimicos a distancia. Patologia 24:225-229,1991
9. Ferrer-Roca O. Narrow and Broad band communications for Telepathology. Tutorials of Cytology Wied et al Ed. Chicago 1994 pp:284-289.
10. Ferrer-Roca O., Ramos A., Diaz-Cardama A. Inmunohistochemical correlation of steroid receptors and disease-free interval in 206 consecutive cases of breast cancer: validation of telequantification based on global scene segmentation. Anal.Cell Pathol 9: 151-163,1995

Back to the T.O.C.