Machine Learning Applications to the Diagnosis of Neurodegenerative Diseases

Introduction 

Imagine you are enjoying a game of Pictionary with your family. As the picturist, you pick up a card from the deck. The card reads “umbrella” as you flip it over. You quickly start sketching an umbrella as the sand timer begins its one minute countdown. As you draw, a family member analyzes the drawing to guess the word. This game of Pictionary is analogous to machine learning, which is a type of artificial intelligence. Artificial intelligence (AI) is broadly defined as the use of computer algorithms in a way that imitates critical analysis and thinking analogous to humans. Machine learning is a subset of AI that allows computer algorithms to make accurate predictions based on a set of data. As children, we are shown pictures of objects, including umbrellas, and are taught that the image of an umbrella correlates to the word  umbrella. This is the process of learning. Having seen umbrellas multiple times, our brains learn to associate the image with the word and can now recognize umbrellas. Similar to how our brains learn, machine learning allows for a set of computer algorithms (also known as a model) to learn by being shown a set of data and taught the patterns among it. The model can then make predictions based on a new set of data by applying the patterns it learned [1]. 

As artificial intelligence (AI) improves efficiency and accuracy, it is emerging as a powerful tool to aid in providing solutions in multiple complex fields. Medicine is an example of a field that AI is used for, particularly the areas of diagnosis and treatment [2]. Since neurodegenerative diseases at present have no cures, early diagnosis and avoiding misdiagnosis are crucial to ensuring patients have a good quality of life [3]. This article will investigate the application of machine learning techniques to the diagnosis and treatment planning of neurodegenerative diseases. 

Machine Learning Models 

A machine learning model is a combination of input data and the consequent output predictions that come with it. Methods of machine learning can be classified into three different types of learning: unsupervised, reinforcement, and supervised [4]. Unsupervised learning models learn by themselves. They are not provided input data that already have output predictions, and instead learn to identify patterns in a data set on their own.  Supervised learning algorithms are more suitable for data pertaining to neurodegenerative diseases. Reinforcement learning is the third type of machine learning, and this type of learning that learns to distinguish between correct and incorrect predictions as it is fed more data. In supervised learning, a model learns over time by being fed input data, known as training data, that already has the correct output predictions. This allows the model to train until it can identify patterns and relationships between input data and pre-fed output predictions, enabling it to make accurate predictions about foreign, new data. Supervised learning is often used for image classification and can be further categorized into three different types: classification and regression. Classification is used when data sets yield qualitative outputs that can be categorized into specific classes [5]. For example, classification algorithms can predict if an image shows a soccer ball or tennis ball based on a given set of images. Regression is used when data sets yield quantitative outputs that are continuous (i.e., can take on a range of values) [5]. For example, regression algorithms could predict the price of a home given the previous sales of homes in the area.

Differentiation of Neurodegenerative Diseases 

Neurodegenerative diseases cause a continuous decline of neural function in the nervous system [6]. The different types of neurodegenerative diseases can be differentiated by examining three key characteristics: location, type, and modification of pathological protein aggregates [6]. Diseases can be identified by the areas of the brain that show degeneration. For example, the basal ganglia and cerebral cortex regions degenerate in Huntington’s disease, and the cerebral cortex region degenerates in Alzheimer's disease [7]. Protein aggregation (the accumulation of misfolded proteins) is a major cause for neurodegenerative diseases, and therefore diseases can be differentiated by the unique roles of biochemical changes involved in protein aggregation that they each possess [8]. Lastly, misfolded proteins accumulate in different cellular locations. For example, misfolded proteins characteristic  of amyotrophic lateral sclerosis are found in the cytosol [7]. 

Diagnosis of Neurodegenerative Diseases through Neuroimaging 

A tool that has proven important in the diagnosis of  neurodegenerative diseases is a neurological screening technique called neuroimaging. Neurodegenerative diseases are mainly caused by aggregation of pathogenic proteins in the brain. As mentioned previously,  the particular type of protein that aggregates helps differentiate different types of neurodegenerative diseases from one another [9]. Examples of pathogenic proteins in neurodegenerative diseases are TDP-43 in amyotrophic lateral sclerosis (ALS), and beta-amyloid & tau in Alzheimer’s disease (AD). 

Given the nature of neurodegenerative diseases, a conclusive diagnosis of the same is often only able to be made post mortem [9]. This is because diagnoses of neurodegenerative diseases requires examination of protein aggregate structures (e.g. senile fibrils in AD) within the brain by autopsy histology, which entails examining cross sections of brain tissue [10].  However, thanks to advances in technology, neuroimaging is now also used to diagnose living patients.

 Neuroimaging can be split into two groups: structural and functional. Structural neuroimaging aids in understanding anatomical brain structure changes, such as loss of brain volume. On the other hand, functional imaging assists in understanding the brain's metabolic changes, such as impaired cellular energy production [11]. Common neuroimaging techniques used in the diagnosis of Alzheimer’s disease (AD) include magnetic resonance imaging (MRI) and positron emission tomography (PET). MRI is used most often to understand changes in brain structure. In Alzhiemer’s disease, MRI imaging is utilized to show brain tissue damage and loss in certain brain regions including the entorhinal cortex and hippocampus, brain volume changes, and rates of atrophy. Brain atrophy is characterized as a loss of neurons and aids in estimation of disease progression [12]. Brain metabolism is sustained through supply of oxygen and glucose to neurons and astrocytes. PET is used to understand functional changes in brain metabolism. Brain metabolism is sustained through supply of oxygen and glucose to neurons and astrocytes. In Alzheimer's disease, PET imaging uses amyloid tracers to estimate amyloid aggregation, which is the causative factor for neurodegeneration in AD patients.In Alzhiemer’s disease, PET imaging estimates amyloid protein aggregation, which is suggested to be one of the main causes of AD, by using amyloid tracers. [13].

Applications of Machine Learning to Neuroimaging 

As mentioned before, neuroimaging is an important tool used in the diagnosis of neurodegenerative diseases. Machine learning can be applied to improve the efficiency of neuroimaging used in disease diagnosis. Specifically, supervised machine learning is most frequently used with medical neuroimaging data. Supervised learning is applied for the purpose of classification to assign and organize a set of data into different categories. There are four types of supervised learning algorithms used in neuroimaging: support vector machine (SVM), decision tree, random forest, and artificial neural networks, with SVM being the most commonly employed type [14]. A support vector machine is a model that specializes in dividing points in a data set into two groups. 

Support Vector Machine Models for Diagnosis of Alzheimer's Disease

SVM models have been applied to classify a group of individuals as those with Alzheimer's disease (AD) and healthy individuals without AD. Magnetic resonance imaging (MRI) is the most common type of neuroimaging associated with diagnosis of AD [15]. Evaluation of MRI images shows brain features such as hippocampus, amygdala, and  grey matter shape and volume and cortical thickness in both left and right brain hemispheres. These brain image features captured from MRI image data were used as input features for the SVM. The SVM model was trained using these input features which are used to identify individuals with AD [15]. The efficiency of the SVM was analyzed by creating a receiver operating characteristic (ROC) curve and calculating the area under the curve (AUC). An ROC curve is a graph plot that determines the prediction and classification accuracy of a machine learning model. The SVM yielded a high AUC value of 0.8606, which proves the SVM had high classification accuracy [15]. 

Conclusion

In the past few decades, artificial intelligence (AI) technology has exploded in popularity. AI’s valuable ability to improve efficiency and accuracy has helped transform numerous disciplines of study, medical diagnosis in particular. Early diagnosis is a crucial step in the management of neurodegenerative disease, given that these diseases devastatingly have no current cures. Machine learning shows to be a promising solution for the issue of early diagnosis of neurodegenerative diseases. Machine learning just may be the beacon of hope for countless patients and their families who strive for a better quality of life.  

References:

[1] Nichols, J. A., Herbert Chan, H. W., & Baker, M. A. B. (2019, February). Machine learning: Applications of artificial intelligence to imaging and diagnosis. Biophysical reviews. Retrieved September 15, 2021, from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6381354/. 

[2] Davenport, T., & Kalakota, R. (2019, June). The potential for artificial intelligence in healthcare. Future healthcare journal. Retrieved September 15, 2021, from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6616181/.

[3] Durães, F., Pinto, M., & Sousa, E. (2018, May). Old drugs as new treatments for neurodegenerative diseases. Pharmaceuticals (Basel, Switzerland). Retrieved September 15, 2021, from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6027455/#:~:text=Currently%2C%20no%20neurodegenerative%20disease%20is,the%20progression%20of%20the%20disease

[4] Myszczynska, M. A., Ferraiuolo, L., Holbrook, J. D., Hautbergue, G. M., Mead, R., Saffari, A., Neil, D., Lacoste, A. M., & Ojamies, P. N. (2020, July). Applications of machine learning to diagnosis and treatment of Neurodegenerative Diseases. Nature reviews. Neurology. Retrieved October 24, 2021, from https://pubmed.ncbi.nlm.nih.gov/32669685/. 

[5] Sidey-Gibbons, J. A. M., & Sidey-Gibbons, C. J. (2019, March). Machine learning in medicine: A practical introduction. BMC medical research methodology. Retrieved October 24, 2021, from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6425557/. 

[6] Dugger, B. N., & Dickson, D. W. (2017, July). Pathology of neurodegenerative diseases. Cold Spring Harbor perspectives in biology. Retrieved September 30, 2021, from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5495060/.

[7] Poddar, M. K., Chakraborty, A., & Banerjee, S. (2021, January). Neurodegeneration: Diagnosis, prevention, and therapy. IntechOpen. Retrieved September 30, 2021, from https://www.intechopen.com/chapters/74857.

[8] Tutar, Y., Özgür Aykut, & Tutar Lütfi. (2013, May). Role of protein aggregation in Neurodegenerative Diseases. IntechOpen. Retrieved September 30, 2021, from https://www.intechopen.com/chapters/44540. 

[9] Shimizu, S., Hirose, D., Hatanaka, H., Takenoshita, N., Kaneko, Y., Ogawa, Y., Sakurai, H., & Hanyu, H. (2018, April). Role of neuroimaging as a biomarker for neurodegenerative diseases. Frontiers in neurology. Retrieved October 24, 2021, from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5915477/. 

[10] Perl, D. P. (2010, August). Neuropathology of Alzheimer's disease. The Mount Sinai Journal of Medicine, New York. Retrieved October 24, 2021, from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2918894/. 

[11]  Szymański, P., Markowicz, M., Janik, A., Ciesielski, M., & Mikiciuk-Olasik, E. (2010, December). Neuroimaging diagnosis in neurodegenerative diseases. Nuclear medicine review. Retrieved October 24, 2021, from https://www.researchgate.net/profile/Pawel-Szymanski-7/publication/49679888_Neuroimaging_diagnosis_in_neurodegenerative_diseases/links/0046351b74d8b949e9000000/Neuroimaging-diagnosis-in-neurodegenerative-diseases.pdf. 

[12] Risacher, S. L., & Saykin, A. J. (2013, January). Neuroimaging and other biomarkers for Alzheimer's disease: The changing landscape of early detection. Annual review of clinical psychology. Retrieved October 24, 2021, from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3955298/. 

[13] Johnson, K. A., Fox, N. C., Sperling, R. A., & Klunk, W. E. (2012, April). Brain Imaging in Alzheimer's disease. Cold Spring Harbor perspectives in medicine. Retrieved October 24, 2021, from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3312396/. 

[14] Yin, T., Ma, P., Tian, Z., Xie, K., He, Z., Sun, R., & Zeng, F. (2020, August). Machine learning in neuroimaging: A new approach to understand acupuncture for neuroplasticity. Neural plasticity. Retrieved November 17, 2021, from https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7463415/.  

[15] Jongkreangkrai, C., Vichianin, Y., Tocharoenchai, C., Arimura, H (2016, March). Computer-aided classification of Alzheimer's Disease. Nuclear medicine review. Retrieved November 17, 2021, from https://www.researchgate.net/publication/299402498_Computer-aided_classification_of_Alzheimer's_disease_based_on_support_vector_machine_with_combination_of_cerebral_image_features_in_MRI/fulltext/570b550608ae8883a1fc44fa/Computer-aided-classification-of-Alzheimers-disease-based-on-support-vector-machine-with-combination-of-cerebral-image-features-in-MRI.pdf.