DOI 10.24412/cl-37136-2023-1-102-106
BIOPHOTONICS AND ARTIFICIAL INTELLIGENCE FOR IMPROVED DIAGNOSTICS: APPLICATIONS IN HEALTH, PHARMACEUTICALS AND AGRICULTURE
PATIENCE MTHUNZI-KUFA123, NKAPHE TSEBESEBE12, LEBOGANG THOBAKGALE1, KELVIN MPOFU1 AND SIPHUMELELE NDLOVU1
1Council for Scientific and Industrial Research, Photonics Centre, South Africa.
2Department of Human Biology, Division of Biomedical Engineering, University of Cape Town, South
Africa.
3School of Chemistry and Physics, University of KwaZulu-Natal. [email protected]
ABSTRACT
A lot of individuals residing in resource limited settings where timely access to medical care is a challenge and healthcare infrastructure is usually poor have no access to laboratory facilities. Disease diagnosis in such sites is dependent on the presence of point-of-care (POC) devices. These POC diagnostics play a key role in ensuring rapid patient care because they are simple to use, inexpensive, portable, instrument independent and do not require a trained technician to operate. To date, optical spectroscopy has become the most important and promising approach in every field of medical, analytical, life, food, and pharmaceutical sciences. However, most currently available spectrometers are bulky and cannot be easily integrated in miniaturized devices. In this work, different biophotonics techniques, artificial intelligence and machine learning models are used to expedite POC diagnostic apparatus for different pathogens, non-communicable diseases, substandard drugs and towards application in food safety as well as security.
INTRODUCTION
Lately, there has been a growing interest in incorporating artificial intelligence and machine learning techniques in biophotonics (1). Biophotonics is an interdisciplinary field that combines the principles of optics, photonics, and biology to study biological systems at the molecular, cellular, and tissue levels. It has a wide range of biomedical applications, including medical diagnostics, drug discovery, and tissue engineering (2). The use of light-based technologies in biophotonics allows for non-invasive, label-free, and high-resolution imaging of biological samples. One of the key advantages of biophotonics is its ability to provide label-free and highresolution imaging of biological samples (3). This allows the study of biological structures and processes without the need for exogenous labels or dyes, which can alter the natural behavior of the sample. Additionally, the use of light-based technologies in biophotonics allows for non-invasive and real-time monitoring of biological systems (2).
On the other hand, machine learning is proving to have great potential as a tool for disease diagnostics (4). Classical machine learning (ML) has proven to be highly effective in various tasks, including classification, regression, and clustering (5). However, there are limitations to classical algorithms, particularly when handling large and complex datasets. This necessitates the development of quantum algorithms (6). Quantum-enhanced machine learning (QEML) can provide significant improvements in the speed and accuracy of ML algorithms for specific tasks. Quantum machine learning offers more efficient and faster computational capacity, making it advantageous when dealing with complex and high-dimensional data. Furthermore, quantum computers can encode complex feature representations that surpass the capabilities of classical computers. Quantum optimization algorithms also offer advantages over their classical counterparts in parameter optimization.
Separately, the growing prevalence of HIV around the globe has continued to promote the growth of the HIV diagnostic market. Sub-Saharan Africa remains the most severely affected region where 70% of HIV positive people in the world reside (7). The global HIV diagnostic market by region comprises of North America, Acia Pacific, Europe, South Africa, and the rest of the world. Tuberculosis (TB) an airborne infection caused by mycobacterium tuberculosis (MTB), until the coronavirus (COVID-19) pandemic, TB was the
leading cause of death from a single infectious agent, ranking above human immunodeficiency virus, an acquired immunodeficiency syndrome (HIV/AIDS) (8). In other reports, the production of substandard medication has become a growing concern in the pharmaceutical industry due its infiltration across all the continents, reaching a billion-dollar estimated value by the World Health Organization (WHO). They also report that the lack of knowledge pertaining to the formulation and distribution of these drugs, affects third world countries who are burdened with various health diseases (9). In general, the most basic way to probe pharmaceutical drugs is by focusing on their chemical behaviour in terms of reaction kinetics and molecular bonding (10-12). Many re-search methods have been proposed and applied for quality control of drugs and such include subdisciplines of chromatography, mass spectrometry and electrochemistry (13-15). Although these methods are successful in detecting and analysing pharmaceutical drugs, their application in large scale drug monitoring and quality control remains limited due to their inherent disadvantages such as complex preparation steps, expensive reagents, long processing times and sample destruction (16-17).
Amidst the modern and complex solutions discussed earlier, it often slipped our minds that optics and photonics can be readily integrated into the field of agriculture. The simplest examples would be the adjustment of plantation direction for optimum sunlight exposure, as well as the usage of incandescent light bulbs in egg incubation and hatching (18). Over recent decades, academics have been alerted to the potential of optics and photonics in the agricultural industry. This has led to progressive developments that utilize optics and photonic techniques in maximizing the quality and productivity of agricultural products.
METHODS
Different optical, photonics and spectroscopy methods are applied in these studies, detailed laboratory protocols can be found in our published literature. Figure 1 (adopted from (19)), below is an example of some of the custom-built optical setups applied in the diagnosis of different microorganisms, work that is intended for point of care diagnostics.
Figure 1: The biosensing rig applied in measuring transmitted light. The system is based on a 512nm green light source with a power of 3.1 mW, collimating lenses (L1 and 2), XYZ stage, a 10x microscope objective, and an imaging system consisting of a CCD camera and computer, focussing lens (L3), an optical fiber, and a portable USB spectrometer connected to a computer (19).
Whilst the current setup is a lab-based tool towards diagnosing communicable with potential applications in assessing markers for non-communicable diseases, the eventual aim is to miniaturize the setup for applications in POC settings.
The field of biophotonics has seen rapid advances in recent years, with new applications being developed all the time. Machine learning techniques have played a significant role in these advancements, as they have the potential to improve the performance and accuracy of biophotonics systems. As machine learning algorithms continue to evolve, the potential for further improvements in biophotonics is expected to continue to grow. Of note, ML algorithms, such as CNNs, have been used to automate the process of image segmentation, which involves separating the foreground of an image from the background. This is an important step in image analysis as it allows for the study of specific regions of interest within an image. In biophotonics, image segmentation has been used to identify and isolate individual cells or structures within a tissue sample. In object detection, ML algorithms have been used to automate the process of identifying and locating objects within an image. This is important in biophotonics as it allows for the identification and tracking of specific cells or structures within a tissue sample. Then in image classification, ML algorithms, such as CNNs and SVMs, have been used to classify images into different categories, such as normal vs. abnormal tissue.
This is important in biophotonics as it allows for the automated diagnosis of diseases such as cancer based on the analysis of microscopic images. Additionally in cell tracking, ML algorithms have been used to automate the process of tracking the movement of cells over time. This is important in biophotonics as it allows for the study of dynamic biological processes, such as cell migration, and can provide insight into the mechanisms of disease (20). Finally in spectral imaging, Machine learning algorithms have been used to process and analyze spectral images, which can contain large amounts of data. Spectral imaging can provide information about the molecular and cellular structure of a tissue sample, and machine learning can be used to extract this information and classify the tissue as normal or abnormal.
The work here follows supervised learning where the algorithms (a convolutional neural network, a support vector machine and a decision tree) are provided with labeled sample data for training to predict outputs. The algorithms are optimized by employing hyperparameter optimization techniques. The techniques were implemented using Python programming language on Google Colab (21-22). The aim of the work is to develop an optimized automatic TB detection system to classify chest X-ray images into healthy and TB infected. The research methodology follows these steps: dataset preparation, hyperparameter optimization, convenient model, comparison of the optimized models and comparisons of the models under study with different lightweight deep learning systems using different performance metrics. The process of classifying chest X-ray images is shown in Figure 2.
Optimized Models •CNN
•SVM
• Decision Tree
Chest X-ray images
Figure 2: An illustration of the approach followed in our research. The input chest X-ray images were adopted from publicly available Kaggle datasets (23-24). The input images are enhanced with image processing techniques (image reshaping, flattening and normalization) to
' ' I
^ ! Pre-processing 1 i !.....................!
minimize error rates and improve detection performance. The output of the preprocessing stage are updated image inputs (25). Thereafter optimized models perform the classification of the images into healthy or tuberculosis, for which their accuracy is evaluated.
CONCLUSION
With the rapid advances in the field, new applications are being developed all the time, and the potential for machine learning to improve the performance and accuracy of biophotonics systems is expected to continue to grow. With the increasing demand for high-precision and real-time data analysis, the use of machine learning in biophotonics will become increasingly important. Continued research and development in this field, allows anticipation to new breakthroughs and advancements in biophotonics.
REFERENCES
[1] P. Pradhan, S. Guo, O. Ryabchykov, J. Popp, and T. W. Bocklitz, Deep learning a boon for biophotonics? Journal of Biophotonics, 13(6). https://doi.org/10.1002/jbio.201960186, 2020.
[2] E. Spyratou, M. Makropoulou, E. Mourelatou, and C. Demetzos, Biophotonic techniques for manipulation and characterization of drug delivery nanosystems in cancer therapy. Cancer Letters, 327(1-2), 111-122, https://doi.org/10.1016/j.canlet.2011.12.039, 2012.
[3] S. Melzer, R. Ankri, D. Fixler, and A. Tarnok, Nanoparticle uptake by macrophages in vulnerable plaques for atherosclerosis diagnosis. Journal of Biophotonics, 8 (11-12), 871-883. https://doi.org/10.1002/jbio.201500114, 2015.
[4] E. Valla, S. Nomm, K. Medijainen, P. Taba, and A. Toomela, "Tremor-related feature engineering for machine learning based Parkinson's disease diagnostics," Biomed Signal Process Control, vol. 75, doi: 10.1016/j .bspc.2022.103551, 2022.
[5] L. Zhou, S. Pan, J. Wang, and A. V Vasilakos, "Machine learning on big data: Opportunities and challenges," Neurocomputing, vol. 237, pp. 350-361, 2017.
[6] N. G. Paterakis, "Hybrid quantum-classical multi-cut Benders approach with a power system application," Comput Chem Eng, vol. 172, 2023, doi: 10.1016/j.compchemeng, 108161, 2023.
[7] A. Buve, K. Bishikwabo-Nsarhaza and G. Mutangadura, The spread and effect of HIV-1 infection in sub-Saharan Africa, The Lancet, 359, 2011-17, 2002.
[8] World Health Organisation, WHO Global Tuberculosis Report, 2022.
[9] https://www.who.int/news-room/fact-sheets/detail/substandard-and-falsified-medical-products
[10] A. Jouyban and E. Rahimpour, Sensors/nanosensors based on up conversion materials for the determination of pharmaceuticals and biomolecules, Talanta., 121383. https://doi.org/ 10.1016/j .talanta, 121383, 2020.
[11] J. Izo, S. Jesus, R. De, Lobenberg and NA, Bou-chacra, Raman Spectroscopy for Quantitative Analysis in the Pharmaceutical Industry. J Pharm.Sci., 24-46. DOI: 10.18433/jpps30649, 2022.
[12] S. Adomaviciute, M. Velicka, V. Sablinskas, Detection of aspirin traces in blood by means of surface-enhanced Raman scattering spectroscopy. J Raman Spectrosc.; 51(6):919-31. https://doi.org/10.1002/jrs.5853, 2020.
[13] O. Heydari Shayesteh, R. Mahjub, A. Ranjbar, K Derakhshandeh and M. Jamshidi M, Nano optical and electrochemical sensors, and biosensors for detection of narrow therapeutic index drugs, Microchimica Acta 188(12):411 Available from: https://doi.org/10.1007/s00604-021-05003-, 2021.
[14] S. Er, U. Laraib, R. Arshad, S. Sargazi, A. Rahdar and S. Pandey, Amino Acids, Peptides, and Proteins: Implications for Nano-technological Applications in Biosensing and Drug / Gene Delivery, Nanomaterials (Basel); 1-36. doi: 10.3390/nano11113002, 2021.
[15] E.C. Le Ru and P.G. Etchegoin, Single-Molecule Surface-Enhanced Raman Spectroscopy, Annu Rev Phys Chem; 63(1):65-87. doi: 10.1146/annurev-physchem-032511-143757, 2012.
[16] I. Aretz and D. Meierhofer, Advantages and Pitfalls of Mass Spectrometry Based Metabolome Profiling in Systems Biology. Int J Mol Sci., 17(5):632. doi: 10.3390/ijms17050632, 2016.
[17] D. Sadutto and Y. Picó, Sample Preparation to Determine Pharmaceutical and Personal Care Products in an All-Water Matrix: Solid Phase Extraction. Molecules, 25(21):5204. doi: 10.3390/molecules25215204, 2020.
[18] S. Sumriddetchkajorn, How Optics and Photonics is Simply Applied in Agriculture? Proceedings of the International Conference on Photonics Solutions (ICPS), Pattaya City, Thailand, p. 888311, 2013.
[19] C. Maphanga, S. Manoto, S. Ombinda-lemboumba, Y. Ismail and P. Mthunzi-Kufa, Localized surface plasmon resonance biosensing of Mycobacterium tuberculosis biomarker for TB diagnosis, Sens. Bio-Sensing Res. 39, 2023.
[20] S. Corsetti and K. Dholakia, Optical manipulation: advances for biophotonics in the 21st century, Journal of Biomedical Optics, 26 (07). https://doi.org/10.1117/1.jbo.26.7.070602, 2021.
[21] E. Bisong and E. Bisong, Google colaboratory. Building machine learning and deep learning models on google cloud platform: a comprehensive guide for beginners, pp.59-64, 2019.
[22] T.B. Kimber, Y. Chen, and A. Volkamer, Deep learning in virtual screening: recent applications and developments. International Journal of Molecular Sciences, vol. 22, pp.4435, 2021.
[23] L. Quaranta, F. Calefato and F. Lanubile, KGTorrent: A dataset of python Jupyter notebooks from kaggle., IEEE/ACM 18th International Conference on Mining Software Repositories (MSR), pp.550-554, 2021.
[24] M.D. Kohli, R.M. Summers and J.R. Geis, 2017. Medical image data and datasets in the era of machine learning - whitepaper from the MIMI meeting dataset session, Journal of digital imaging, vol. 30, pp.392-399, 2016.
[25] S.I. Nafisah, and G. Muhammad, Tuberculosis detection in chest radiograph using convolutional neural network architecture and explainable artificial intelligence, Neural Computing and Applications, pp.1-21, 2022.