Machine Learning: Train a Machine to Do Your Job (some of your job)

Author: Alexandra Davis

Synopsis: Machine (model or classifier) learning (ML) is a subfield of artificial intelligence and is behind many technologies we use on a daily basis. Can we use them in clinical practice? Imagine a chatbot can give therapeutic advice; a text completion function we use in emails can incorporate test data and finish the rest of the paragraph, or a suggestion algorithm comes up with possible diagnoses after reading the patient’s data. The potential to utilize machine learning in the field of neuropsychology is limitless.

Author Disclosures: Nothing to disclose

Overall Description:

1. Clinical decision support and predictive analytics: A trained model can predict a patient’s risk level or prognosis for specific neurological, cognitive, or psychiatric disorders, which assists with diagnostic decision-making and treatment intervention.
2. Identifying diseases and diagnosis: Machine learning models have been used to diagnose heart disease, diabetes, cancer, etc. One ML model uses specific gait characteristics to differentiate Parkinson’s Disease from other neurological diseases (Amyotrophic lateral sclerosis and Huntington’s Disease).
3. Natural language processing (NLP):  NLP enables computers to understand natural language as humans do. An NLP model could analyze research articles and identify all relevant diagnostic markers or treatment strategies.
4. Machine Learning-based Behavioral Modification: Many therapeutic goals are focused on behaviors such as quitting smoking, reducing alcohol consumption, eating a healthier diet, or becoming more physically active. ML models can analyze data from sensors and wearable tech to fine-tune treatment and reduce damaging behaviors.

Case Study Example: SVM Based Machine Learning Approach to Identify Parkinson’s Disease Using Gait Analysis

Authors measured the gait of patients with Parkinson’s disease (n = 15), Huntington’s disease (n = 20), Amyotrophic lateral sclerosis (n = 13) with stride, swing, and stance, etc. The data set contains 12 features related to the human gait cycle. Features were then reduced using a correlation matrix. These feature vectors are then individually analyzed to extract the best 7 feature vectors which are then classified using a Gaussian radial basis function kernel-based Support vector machine (SVM) classifier. Results show that the 7 features selected for SVM achieve a good overall accuracy of 83.33%, a good detection rate for Parkinson’s disease of 75%, and low false positive results of 16.67%.

Limitations:

A major limitation is that ML typically needs more than 1000 data points to sufficiently train a model Therefore,  diseases with low prevalence may not be appropriate. It is also hard to trust the result from an ML model if we don’t understand how it came about. Overfitting can also become a problem, where the evaluation of ML algorithms on training data differs from test data.  This happens when the model does not categorize the data correctly, it learns from the noise and inaccurate data entries and results in a high variance in test data.

Justice, Equity, Diversity, and Inclusion Issues:

Machine learning models are only as good as the data used to train them. If the data used to train the models does not represent the population you are trying to serve, it can lead to biased results. Representation of minority groups within datasets can be limited, leading to under-representation in machine learning models.

Helpful links to further reading/material and references

Shetty and Y. S. Rao, “SVM based machine learning approach to identify Parkinson’s disease using gait analysis,” 2016 International Conference on Inventive Computation Technologies (ICICT), 2016, pp. 1-5, doi: 10.1109/INVENTIVE.2016.7824836. https://ieeexplore.ieee.org/abstract/document/7824836

An introduction to Machine Learning: https://www.geeksforgeeks.org/introduction-machine-learning/

A Beginner’s Guide to Neural Networks and Deep Learning: https://wiki.pathmind.com/neural-network

The Limitations of Machine Learning https://towardsdatascience.com/the-limitations-of-machine-learning-a00e0c3040c6

Glossary/definitions:

Natural Language Processing (NLP): a computer program’s ability to understand human language as it is spoken and written — referred to as natural language. It is a component of artificial intelligence (AI). https://www.techtarget.com/searchenterpriseai/definition/natural-language-processing-NLP

Machine Learning (ML): is a subfield of artificial intelligence that gives computers the ability to learn without explicitly being programmed. https://mitsloan.mit.edu/ideas-made-to-matter/machine-learning-explained

Model: A model is a specific representation learned from data by applying some machine learning algorithm. A model is also called a hypothesis.

Algorithm: a procedure run on data to create a machine learning “model.”