It occurred to the professors, Jakob J. Stamnes and Knut Stamnes, that the method they developed for satellite remote sensing of the atmosphere-ocean system, may be used also for detection of skin diseases, since skin can be considered as a layered medium in which the optical properties of each layer can be correlated with physiological properties and morphological parameters through a bio-optical model. Further elaboration of this idea led to the Optical Transfer Diagnosis technology for skin cancer detection.
Jakob J. Stamnes, professor emeritus at the University of Bergen, Norway, and Knut Stamnes, professor at Stevens Institute of Technology, USA, have over many years developed models and algorithms for optical satellite surveillance of the ocean. Their model describes how sunlight propagates through the atmosphere into the ocean through the air-ocean interface, and then interacts with oceanic substances by absorption and scattering, after which part of the light is scattered upwards from the ocean and received by detectors on board a satellite deployed in space.
By analysing the upward scattered sunlight from the layered atmosphere-ocean system received by a satellite instrument, one can determine the concentrations of atmospheric constituents as well as algae and other substances in the ocean. It occurred to the two professors that this analytic method may be used also for detection of skin diseases since skin can be considered as a layered medium in which the optical properties of each layer can be correlated with physiological properties and morphological parameters through a bio-optical model. Further elaboration of this idea led to the Optical Transfer Diagnosis technology or DermoSight technology for skin cancer detection.
The DermoSight technology, developed over many years by Balter Medical AS (Balter), Bergen, Norway, is based on illuminating a selected area of the skin (e.g. a healthy mole or suspicious lesion) with light of different colours from different directions and measuring the light backscattered from the lesion in different directions. These measurements are used to infer physiological properties and morphological parameters of the tissue, which differ between benign and malignant tissue. Balter has also developed a prototype DermoSight dermatoscope, which is a spectral radiometer (see Fig. 2) that records a set of 30 images, constituting a lesion measurement, in less than 10 seconds. Images of the lesion are recorded at 10 different wavelengths (365–1,000 nm) from multiple angles of illumination and detection.
The data acquisition geometry of the prototype DermoSight dermatoscope was designed in such a way that for each combination of illumination and detection directions the same area of the skin was interrogated, allowing a one-dimensional treatment when the independent-column approximation is invoked and the skin tissue is assumed to have a layered structure:
- an uppermost layer, the epidermis, consisting of an upper part and a lower part
- the dermis, containing the blood circulation
- the subcutis, a strongly scattering fat-containing layer.
The inherent optical properties of each layer are the absorption and scattering coefficients as well as the scattering phase function (describing the angular variation of the scattered light), each varying with wavelength.
The DermoSight technology is based on light propagation calculations and non-linear inversion, which are used to retrieve maps of five physiology properties:
1. Percentage of hemoglobin concentration
2. Percentage of hemoglobin oxygenation
3. Percentage of melanosome concentration in upper epidermis
4. Percentage of melanosome concentration in lower epidermis
5. Percentage of keratin concentration
And two morphology parameters:
1. Upper epidermal thickness
2. Lower epidermal thickness.
Each of these five physiology properties or two morphology parameters is retrieved pixel by pixel to provide a map covering the lesion and a surrounding area. Although there is no independent verification of any of these retrieved values by application of different methods for measuring them, they have been found useful in an intermediate step leading to a final diagnostic indication.
From each map, an entropy value is calculated and cross entropy values are calculated for different pairs of maps. The entropy concept used here is similar to that used in statistical physics and information theory. These entropy and cross entropy values together with a number of morphometric parameters derived from the nadir green image are used to define a total of 86 diagnostic parameters. For each independent lesion measurement, a diagnostic index is defined as a weighted sum of the diagnostic parameters.
Balter has developed its own clustering technique and applied it for lesion classification in order to obtain discrimination between malignant (class 1) and benign (class 2) lesions through the identification of one set of class 1 clusters and another set of class 2 clusters, each cluster comprising a certain number of independent lesion measurements. Standard methods of clustering were considered but found not to be applicable since it is not known a priori how the diagnostic parameters defined above can be used to discriminate between objects from opposite classes, which are organized in accordance with dermoscopy evaluations and biopsy results.
Balter has successfully developed the DermoSight algorithm for discriminating between benign and malignant lesions using:
- Neural network technology
- Cost function optimization to determine optimal diagnostic weights
- Generalized diagnostic parameters
- Diagnostic indication by a team of experts.
Product development
Balter Medical AS
A part of DeepX Health
Thormøhlensgate 55
5008 Bergen
Norway
Jakob J. Stamnes
jakob.stamnes@baltermedical.com
+47 95 77 00 66