Light transport in three-dimensional plane-parallel tissue slabs has been modelled by Monte Carlo analogue simulation. The model design has allowed the study of transmission properties that are pertinent to imaging systems for the detection of breast cancer. An important aspect of the investigations is that they make use of data obtained from quantitative measurements of light scattering and absorption in normal and pathological breast tissues. It is shown that an imaging technique which used a raster scanning laser and detector arrangement and plane-parallel compression of the breast could have considerable advantages in terms of improved transmittance, spatial unsharpness and contrast. Time-of-flight gating of images is also found to be beneficial provided that the light intensities after temporal filtering remain adequate.

Optical experiments are described for measuring the attenuation characteristics of breast tissues at visible and near-infrared wavelengths. Total attenuation coefficients post mortem were measured directly in thin tissue sections. They are usually within the range from 10 to 30 mm-1, are rather higher in fat than in fibroglandular specimens and decrease with increasing wavelength. The scattering phase function is strongly forward-peaked with the mean cosine of scattering in the range from 0.85 to 0.97 and appearing more forward-peaked in fat than in fibroglandular tissue. The reduced scattering coefficient is of the order of 1 mm-1 in all tissues. Absorption coefficients were measured indirectly in optically thick sections. They are typically between 0.1 and 0.5 mm-1 at wavelengths around 580 nm and an order of magnitude lower at 850 nm. At 580 nm and shorter wavelengths the absorption in carcinoma is significantly higher than in adjacent uninvolved tissue. Significant differences were observed in the first-order derivatives of the transmission spectra of carcinoma and surrounding tissues at certain infrared wavelengths. Transmission spectra measured in vivo across the wavelength range from 500 to 860 nm have a similar form to the spectra of excised samples. Linear absorption coefficients are generally of the same order of magnitude as those found in vitro although they are lower at green wavelengths.

Two-dimensional images obtained using ultrasound have been digitized from videotape recordings and stored within a maximum of 240 digital memory planes to form a three-dimensional data set using a commercially available image processing unit. This data set has been manipulated to produce images in planes perpendicular to the original scan set. The reformatted images represent not only the scans that could have been obtained by rotating the scan head but also demonstrate planes that are not accessible by conventional scanning. The system has been evaluated with a tissue-equivalent phantom to determine the geometric accuracy of the reformatting process. Clinical material has also been used to evaluate the practical value of such a technique and to highlight difficulties that may be encountered in its routine use.

A review of the current state of transillumination imaging for the detection and diagnosis of breast cancer and the difficulties that impede more widespread acceptance of the methods is presented. An outline is given of the physical models that may be used to describe the propagation and scattering of light in a tissue matrix and how these models might be valuable in identifying imaging improvements. Some of the proposals for future imaging arrangements are described and the preliminary work on a system for light transmission computed tomography is presented.

Digital subtraction angiography (DSA) allows the degree of arterial patency or stenosis to be rapidly quantified. We have assessed the accuracy with which a single-plane DSA system is able to quantify area patency by densitometric and geometric methods. Arterial phantoms were designed to test for systematic error; intra-arterial DSA images of critical lesions of the carotid bifurcation and the lower abdominal and peripheral vessels were used to determine intra- and interobserver reproducibility. The densitometric method, which was more accurate than the geometric method, had a mean systematic error of up to 4% and a mean intra-observer variability of about 15% (coefficient of variation). We have identified the principal sources of inaccuracy and ways in which it may be reduced.

The potential and problems of the superimposition of medical images from different techniques is addressed. Examples of current work and methods of combining images from nuclear medicine and radiology are presented, with discussion of their usefulness.

aDepartment of Anesthesiology and Pain Clinic, University Hospital, Leiden, the Netherlands bDepartments of Psychiatry and Pain Clinic, University Hospital, Leiden, the Netherlands