Skip to main content

Web Content Display Web Content Display

SOLARIS centre

Web Content Display Web Content Display

Scientific publications list

Web Content Display Web Content Display

New method of data quality improvement by noise removal

New method of data quality improvement by noise removal

Scientists from the CIRI beamline have found a way to accelerate infrared imaging of biological tissues. Their solution will enable faster diagnosis of cancer thanks to denoising. You can read about the MNF2 method they developed in the journal "Chemometrics and Intelligent Laboratory System". With this discovery, researchers significantly enhanced the measurement capabilities using the FPA array detector available at the FT-IR microscopy end station at the CIRI beamline.

In infrared imaging (IR), a key aspect of the analysis of biological tissues is the time necessary to perform the measurement. This is particularly important in terms of the diagnosis of tumors in tissues - where it can be used as an alternative to conventional histopathological staining as well as support for histopathologists in the diagnosis of the disease. One of the approaches to shorten the measurement time is to reduce the number of measured scans to, e.g. 4 - which are employed in studies on tissue classification - instead of the typically measured 128 or 64 scans. This results in increased noise in the spectra. However, by using data pre-processing methods, specifically denoising, they can obtain data qualitatively comparable to those measured with a large number of scans. Nowadays, the method gaining popularity among researchers working with IR imaging is the Minimum Noise Fraction (MNF) method. This method is based on the eigenvalue decomposition. In the first step: the noise matrix is calculated and decomposed. This estimate is made by subtracting the signal from neighboring pixels, assuming that they do not differ in terms of chemical composition - the only difference should be noise. However, as researchers showed in their publication, this is only close to the truth in the measurement configuration with the highest spatial resolution. Therefore, they proposed a solution to this problem (called MNF2) which involves measuring the sample (breast cancer tissue) sequentially, i.e. two measurements with two averaged scans each are performed. This allowed researchers to estimate the noise matrix by subtracting sequential measurements - the signal from exactly the same pixels is subtracted and the noise estimation is now independent of the spatial resolution. The proposed MNF2 method, through a minimal modification of the data collection step, eliminates signal blur during denoising, which improves spatial resolution and allows correct data interpretation. This method significantly increases the measurement capabilities using the FPA array detector available at the FT-IR microscopy end station at the CIRI line. 

Fig. On the left side, a breast needle biopsy imaged with FT-IR is presented, with three interesting tissue regions: necrosis, blood, and fiber. On the right side,  FT-IR spectra are shown (coming from pixels marked with filled squares), for visual comparison of denoising methods performance.

Fig. On the left side, a breast needle biopsy imaged with FT-IR is presented, with three interesting tissue regions: necrosis, blood, and fiber. On the right side,  FT-IR spectra are shown (coming from pixels marked with filled squares), for visual comparison of denoising methods performance.

 

Author: Danuta Liberda-Matyja


Link to the publication: D. Liberda-Matyja, T. Wrobel, New resolution independent approach to noise estimation in Minimum Noise Fraction denoising of tissues measured with Infrared Imaging, Chemometrics and Intelligent Laboratory Systems, 244, 105042(2023) doi: 10.1016/j.chemolab.2023.105042

This research was supported by the NCN grant (2019/35/N/ST4/01809).

Recommended
Hercynite – the mysterious refractory spinel compound understood!

Hercynite – the mysterious refractory spinel compound understood!

Direct observation of the spin texture and the Rashba effect of a ferroelectric semiconductor

Direct observation of the spin texture and the Rashba effect of a ferroelectric semiconductor

Detection of early pancreatic cancer lesions using infrared and machine learning

Detection of early pancreatic cancer lesions using infrared and machine learning

2D boron layer as a nanoscopic diffraction grating

2D boron layer as a nanoscopic diffraction grating