Medical treatment uses tools for working with randomized controlled trial data, where we only observe what happens to a patient if they’ve got treatment, and we want to know what would happen to the patient if they had not received the treatment and vice versa.
Machine learning allows us to build models that associate a broad range of variables with the effect of a treatment on individual patients.
The data science and machine learning platform Neural Designer combines different data types into a single model to better diagnose diseases.
Contents
Objectives
Medical treatment aims to assess which tools better fit each patient to survive the disease. Besides, the objective is to determine one or more treatments, whether we want to avoid acquiring a disease or condition, suffer from symptoms, have caught a cold or the flu, have developed diabetes or cancer, or have injured yourself in an accident. All of these require treatments.
Thankfully, a large amount of data is currently available to clinicians, ranging from details of clinical symptoms to various types of biochemical assays and outputs of imaging devices.
We put up a model to predict the best type of treatment. These are some examples of types of data that could be useful to make a proper medical treatment using machine learning:
- Personal data: Age, sex…
- Disease data: Physiological measurements and information about known diseases or symptoms that an individual has experienced.
- Environmental data: Information about an individual’s environmental exposures, such as smoking, sunbathing, weather conditions, etc.
- Genetic data: All or key parts of the DNA sequence of an individual.
- Treatment data: Information about different treatments that can be used.
Benefits
Treatments fall into four categories based on their potential outcomes. Some benefits of leveraging machine learning in medicine are the following:
PREVENT
Avoiding a wealth problem before it starts, using vaccines and personalized medicines to prevent genetic diseases.
CURATE
Curing, healing, or repairing after being diagnosed with a disease or condition to return to 100% of our health. For this, we can use different antibiotics or antihistamines. Surgical treatment attempts to repair a problem and physical therapy for muscles and other parts that have been strained and damaged.
DISEASE MANAGEMENT
Existing medical treatments cannot cure many diseases and conditions. The goal here is to control the problem and maximize the patient’s longevity and quality of life.
PALLIATE
Palliative care is provided to patients with severe, debilitating, and life-ending diseases. Its goal is to relieve symptoms such as pain and make a patient comfortable, with little or no attempt to cure or manage the disease or condition that causes the discomfort.
Approach
Neural networks are the most important technique for machine learning and artificial intelligence. The following flow chart shows how to build and use a neural network for medical prognosis.
The first step is to create a data set by collecting all the personal, environmental, disease, and comorbidity data about a cohort and the outcome of those patients.
Then, we need to build a neural network that will forecast the outcome of new patients.
The next graph illustrates a neural network for medical prognosis.
A training strategy is applied to the neural network to discover the underlying relationships in the data set.
To improve the model’s predictive capabilities, we can also apply model selection by trying combinations of variables and choosing those with more impact on the outcome.
Then, the resulting model undergoes an exhaustive testing analysis.
Finally, after model deployment, the neural network is used to predict the future health of new patients.
The data science and machine learning platform Neural Designer guides you through this process to focus on the medical aspects and not on the details behind machine learning.
Conclusions
Machine learning enables us to build predictive models that help doctors predict the outcome of a disease and choose the best possible treatment for each patient.
References
- Kononenko, I. (2001). Machine learning for medical diagnosis: history, state of the art and perspective. Artificial Intelligence in Medicine, 23(1), 89-109.
- Gagliano, M., Van Pham, J., Tang, B., Kashif, H., & Ban, J. Applications of Machine Learning in Medical Diagnosis.
