Identification of spectral features differentiating fungal strains in infrared absorption spectroscopic images

Abstract

There are many unknowns regarding the interaction between fungi and their surroundings. In this project, we took a closer look at hyperspectral images of several fungal strains on two different substrates. The project mainly consisted of developing a code for the classification of fungal strains and the extraction of information from it. The classifier used hyperspectral images in infrared of four strains of three species that were grown on two different substrates. A total of 192 images were used. Images were processed using software that was already created for the analysis of hyperspectral data. We developed a random forest classifier to classify the samples by fungal strains. The performance of different classifier parameters was determined and the best ones were chosen. Then, spectra and their derivatives were analyzed and their classification performances were compared. As the last step of the project, the developed random forest algorithm was used to identify the most important wavenumbers for discerning different fungal strains. One of the interesting results was an unexpectedly high increase in the accuracy of the classifier when the first derivative of spectra was used instead of plain spectra.

Popular Abstract

Fungi are all around us and as such are used in many industries (e.g.\ farming and medicine). One example is fungi that play a crucial role in tree growth and development. Actually, a tree's root system and a fungus form a strong relationship that is an example of a symbiosis. The fungus supplies the tree with nutrients from the ground, while in return the tree provides carbohydrates obtained through a process of photosynthesis to the fungus. The prime example of this type of symbiosis is the ``humongous fungus", a fungus that interconnects the whole Malheur National Forest in Oregon. Grasping the interconnectedness between trees and fungi could teach us more about what we can do to create healthier forests. Healthy forests are of foremost importance, especially now when we are facing an unprecedented number of wildfires and ever-increasing pollution of the air.


Currently, we are not sure exactly how fungi interact with the soil surrounding them. In order to find out, we need to observe a region around a cell wall with sufficiently high spatial resolution. For years it was hard to imagine having fungi on a substrate that is nice enough for imaging and at the same time not harmful for the organism. In recent years that became possible. Still, to get a high spatial resolution image at a certain wavelengths of light takes too much time which makes taking the whole spectrum at high resolution infeasible. This problem is circumvented by taking cruder images for many wavelengths, determining which ones are interesting for further investigation, and then using a higher resolution imaging technique.

One of the main goals of this project was to analyze spectroscopic data of several fungal strains. A machine learning algorithm was used to deduce the most important wavelengths for differentiating among the strains. We hope that those wavelengths will prove useful for future high-resolution spectroscopic analyses.

• author: Stancevic, Dejan
• supervisor: Carl Troein
• organization: Computational Biology and Biological Physics - Undergoing reorganization
• course: FYTK02 20212
• year: 2022
• type: M2 - Bachelor Degree
• subject: Physics and Astronomy
• language: English
• id: 9075875
• date added to LUP: 2022-03-14 16:27:35
• date last changed: 2022-03-14 16:27:35