Skin cancer is the most common type of cancer in the world, requiring tens of thousands of surgical biopsies of suspicious lesions. According to the World Health Organization, two to three million non-melanoma and 132,000 melanoma skin cancers are diagnosed globally each year.

A noninvasive tool that can distinguish benign from malignant cutaneous lesions could help limit the need for biopsy to only highly suspicious lesions. This could reduce the number of biopsies performed and potentially significantly reduce the cost of skin cancer diagnosis.

Researchers from Spain have now developed a 3D scanner, based on fringe projection and machine learning, that shows potential as an in vivo skin cancer detection device. The prototype system obtains morphological parameters of skin lesions related to area, volume and perimeter with micrometric precision and can distinguish between melanomas and moles. By quantifying the volume and shape of the lesion, its capabilities extend beyond the conventional qualitative palpation used by dermatologists (Biomed. Opt. Express 10.1364/BOE.10.003404).

Fringe projection can be used to acquire quantitative information on surface heights of skin in several seconds, creating a height map that can be used to help detect non-melanoma skin cancers. High-resolution digital cameras, economically priced real-time frame grabbers and powerful image processing software have made this a viable technology to analyse skin topography.

Led by Santiago Royo and Meritxell Vilaseca Ricart, a team of researchers at the Universitat Politècnica de Catalunya designed a compact handheld prototype incorporating two monochrome CCD cameras and a Pek3 picoprojector. These create an image of a fringe pattern on the skin, which is moved horizontally during acquisition. Upon completion, the patient’s skin is illuminated with a uniform white field and a colour image is acquired. The three camera images are then calibrated, and undergo image processing to create point-wise height maps. A machine learning-based classification system then identifies the types of lesions.

Clinical testing

For the validation study, the team examined patients with a total of 654 skin lesions who had dermatology evaluations at the Hospital Clinic i Provincial de Barcelona and the Università degli Studi di Modena e Reggio Emilia. Royo told Physics World that the researchers enrolled patients who came for treatment and that they represented an overwhelmingly older Caucasian population.

While 608 lesions were in body locations that could be measured with the 3D scanner, only 194 (32%) could be analysed, due to micro-movements, inaccurate hair removal, and/or locations outside the system’s field-of-view. The authors reported that 43% of these lesions were cancerous: 31% melanomas, 9% basal cell carcinomas and 3% squamous cell carcinomas. The remainder were benign nevi and non-nevi lesions. The scanning system achieved 80.0% sensitivity and 76.7% specificity.

“In our pilot, we analysed all suspect lesions, which were subsequently confirmed by histology. It was unfortunate that there were so few squamous cell and basal cell carcinomas in this population,” says Royo. “It would be interesting to add more of these lesions to improve the specificity regarding these two cases, and to better separate them from melanoma. We are currently applying for funds to conduct this type of research.”

The team also hopes to extend the research to evaluate skin lesions on persons of colour, both African and Asian, which could be achieved by adjusting the intensity and contrast of the projected fringes.

“Our next goals are extending the study to improve the training network, and to combine the information with other sensors to get multiple inputs, which include live flow measurements using optical feedback interferometry and multispectral imaging,” says Royo. “We would also like to extend the technology to other diseases, in particular other types of cancer and illnesses related to blood flow abnormalities.”