Development of a new spectral imaging system for the diagnosis of skin cancer
- Autores: Xana Delpueyo Español
- Directores de la Tesis: Meritxell Vilaseca Ricart (dir. tes.), Santiago Royo Royo (codir. tes.)
- Lectura: En la Universitat Politècnica de Catalunya (UPC) ( España ) en 2017
- Idioma: español
- Tribunal Calificador de la Tesis: Maria Cojocaru Crina (presid.), Dragan Indjin (secret.), Jussi Parkkinen (voc.)
- Programa de doctorado: Programa de Doctorado en Ingeniería Óptica por la Universidad Politécnica de Catalunya
- Materias:
- Enlaces
- Tesis en acceso abierto en: TDX
- Resumen
The incidence of skin cancer in Europe, US and Australia has been rising rapidly. Skin cancer accounts for one in three cancers worldwide and a person has 1:25 chance to develop a melanoma, the most aggressive form. Visual inspection followed by histological examination is, still today, the gold standard for clinicians, which is carried out through a dermoscope, a handheld device with a magnifying lens and a white and uniform illumination field. The dermoscopic technique requires considerable training in the interpretation of what is seen and is highly dependent on subjective impressions. In consequence, a large number of unnecessary surgical procedures are performed.
For this reason, in this thesis a spectral imaging system to improve skin cancer diagnosis has been developed. This work has been carried out in the framework of the European project DIAGNOPTICS "Diagnosis of skin cancer using optics", which aimed to launch a hospital service based on a multiphotonic platform to improve skin cancer with the combination of four non-invasive novel techniques: 3D and multispectral imaging, optical feedback interferometry and confocal microscopy.
The handheld system built included a monochromatic CCD camera attached to an objective lens and a light source containing 32 light emitting diodes (LEDs) with 8 spectral bands from 400 nm to 1000 nm. An acquisition software to control all the components of the multispectral system was programmed as well as a simplest version for physicians. The changes over time of the emission of the LEDs was analysed, and also the linear response of the camera at each wavelength, the uniformity of the LED emission and the short and long-term repeatability of the system in acquiring images, to ensure the good performance of the system. In order to proceed with the Ethical Committee approval and to launch the systems in both hospitals, irradiance and radiance measurements were done according to the standard UNE-EN 62471.
A Graphical User Interface (GUI) was developed for the spectral images processing and corresponding analysis, allowing spectral and colorimetric features to be computed in terms of reflectance, absorbance and colour parameters. Furthermore, a segmentation algorithm was also implemented to extract the isolated information from the lesion. For all images calculated in terms of any of the parameters, conventional statistical descriptors were obtained. As a first approach to extracting textural information we also used the analysis of the statistical properties of the histogram.
An inclusion criteria and a measurement protocol were established. From all lesions analysed, 620 were measured with the multispectral system, 572 of them had a clinical or histopathological diagnosis, and 502 could be properly segmented. Therefore, 429 skin lesions were finally included in the study: 290 nevi, 95 melanomas and 44 basal cell carcinomas. A classification algorithm was developed in order to decide whether the lesions were malignant (melanomas and basal cell carcinomas) or not (nevi), splitting previously the data into training and validations set of the same size. 15 parameters from 1309 were found to be not redundant providing a 91.3% of sensitivity and 54.5% of specificity. Accordingly, the addition of textural information was shown to be useful for the diagnosis of malignant lesions than the sole use of averaged spectral and colour information.
The same steps were carried out for the 3D imaging system also included in the multiphotonic platform. In this case, 3 parameters were found to be useful for the classification providing values of 55.6% and 83.7% of sensitivity and specificity, respectively. Finally, the combination of both system was also tested as a first attempt to improve the detection of melanomas, providing 100% and 72.2% of sensitivity and specificity, respectively. However, the conclusions reached in this case should be taken with caution due to the limited number of lesions.
