Image analysis based on an optical model of the skin for detection of early signs of melanoma


Funded by EPSRC, grant number GR/M53035

This research is concerned with the characterisation of pigmented skin lesions to help with early diagnosis of malignant melanoma, a skin cancer. Out group has developed a novel image analysis method which uses physics-based modelling of optical properties of the skin. The method computes parametric maps characterising skin structure and composition. The images show histological quantities in the skin, such as concentration of pigment melanin, concentration of blood and thickness of collagenous tissue. They also show whether melanin is present in the dermis - such presence is a very sensitive indicator of melanoma.

Skin structure

The skin consists of a number of layers with distinct function and distinct optical properties. White light shone onto the skin penetrates superficial skin layers and whilst some of it is absorbed, much is remitted back and can be registered by a camera.

Model of tissue colouration - normal skin

The key to the interpretation of image colours in terms of the underlying histological parameters is a model of tissue colouration which provides a cross-reference between the colour and the histology. This model is constructed by computing the spectral composition of light remitted from the skin given parameters specifying its structure and optical properties. This step needs to be carried out only once. As the mapping between the colours and the parameters is unique for the skin, each colour corresponds to one specific set of histological parameters.

Expressed as a fragment of a pseudocode, the process of building of the model of colouration can be described as follows:

given 
	incident light
	absorption coefficients of melanin and blood
	scatter coefficient of the papillary dermis
	scatter coefficient and thickness of the reticular dermis
	spectral response functions for the red, green blue and nir primaries
for all valid concentrations of epidermal melanin
 for all valid concentrations of dermal blood
  for all valid thicknesses of papillary dermis
    compute
	remitted light spectrum
	colour vector [r  g  b  nir]

This forward process computes explicitly tissue colour given a set of histological parameters. As the mapping between the histological parameters and the primaries is unique for the skin, the inverse mapping is possible: from the tissue colour to its histological parameters:

	[r  g  b  nir] <-> [ melanin   haemoglobin  papillary_dermis ]

The quantities [ melanin haemoglobin papillary_dermis ] are then used to construct parametric maps.

The model of normal skin colouration is representative of all the normal skins, irrespective of racial origin, age or gender. The structure remains the same, and the only differences are in the magnitudes of the parameters.

Abnormal skin colouration

The model above has been constructed for skin which has a normal structure. Skin colouration associated with abnormal conditions does not necessarily have to conform to this model.