Multi-Spectral Optimization for Tissue Probing Using Machine Learning

Yarden Tzabari Kelman, Hadas Lupa Yitzhak, Nadav Shabairou, Shahaf Finder, Zeev Zalevsky

Research output: Contribution to journalArticlepeer-review


An optical approach for pigmented lesions detection in human skin is presented. As differences between normal skin tissue and pigmentation tissue (and even a change in pigmentation development) can be detected by their optical properties, this paper presents a new, potentially noninvasive approach for skin cancer detection. Since each wavelength has different penetration depth into the tissue and different absorption, the goal was to check whether a combination of information obtained from five different wavelengths, can increase the detection probability and reduce the false positive probability, compared to using only one wavelength. Temporal tracking of back-reflected secondary speckle patterns generated while illuminating the tested area with several lasers and applying periodic vibrations to the surface via a controlled vibration source at several stimulation frequencies. As a sequel to the previous work conducted in our laboratory which investigated pigmented lesions interaction with one light source, this work deals with increasing the number of parameters that are being looked at and considered at the same time. Using five wavelengths, 9 vibration frequencies and 19 signal analysis parameters, ex-vivo experiments were performed on porcine skin tissues and were analyzed using artificial intelligence tools which could detect the strong features for each wavelength individually. Combining the wavelengths produced impressive results compared to the results by each wavelength separately: both types of errors, false positive and false negative, decreased to less than 2%. Such a significant change in its impact on patients shows the value of this method. This paper shows the possibility of optically separating normal skin from pigmentation tissue, by using the advantages of multi-spectral optimization. This is a necessary proof of concept as a preliminary step toward our future experiments, which may differentiate between different types of pigmentation, and even malignancy and benign tissues.

Original languageEnglish
Article number9310312
JournalIEEE Photonics Journal
Issue number1
StatePublished - 1 Feb 2021


  • Skin optical properties
  • ex-vivo
  • lasers
  • machine learning
  • optics
  • optimization
  • pigmented lesions
  • speckle
  • tissue probing

All Science Journal Classification (ASJC) codes

  • Atomic and Molecular Physics, and Optics
  • Electrical and Electronic Engineering


Dive into the research topics of 'Multi-Spectral Optimization for Tissue Probing Using Machine Learning'. Together they form a unique fingerprint.

Cite this