Machine learning in the Healthcare Industry

Machine learning (ML) is an integral part of artificial intelligence that has revolutionised a huge array of sectors and industries across the world, including the medical and healthcare industries.

ML has already made remarkable contributions to the field of healthcare – this is a key sector for the application of AI and ML.

Healthcare systems are critical to everyone and they rely on accurate measurement and health diagnostics to create optimal healthcare outcomes.

It is actually one of the first sectors to incorporate AI and machine learning, and in an industry where speed and accuracy are absolutely paramount, AI enhances decision making processes to significantly improve outcomes for both practitioners and patients.

---

The Power Of Artificial Intelligence In The Medical Industry

The introduction of AI in medical facilities has not only enhanced healthcare outcomes but has also drastically reduced the cost of supply and administrative costs. Today, any healthcare system with machine learning has unparalleled control over extensive public-access data libraries.

AI can scan, analyse, and process huge volumes of medical data and provides impressive medical insights on:

• Diagnosis of Diseases.
• Personalised Medicine & Treatment Plans.
• Immunology.
• Modelling the virality of viruses (epidemiology).

---

The Benefits of ML in the Healthcare Industry

ML’s exceptional ability to work efficiently with data makes it suitable for the healthcare industry where there is an abundance of medical practices that are focused on the prognosis, diagnosis, progression and treatment of diseases.

ML Improve’s Operational Efficiency

AI and machine learning is:

• Able to process multiple tasks at once.
• Can make objective decisions more quickly than a human.

Therefore, AI can replace entire teams of clinicians who spend a significant amount of time working on rudimental tasks (e.g. appropriating doses of different drugs, or going through scans, etc). This allows staff to re-focus their time to solving more complicated or valuable problems/tasks.

Highly Accurate Diagnostic Results 

Combined with time efficiency is accuracy. AI has beaten radiographers in spotting lung cancers and eye disease, to name just two conditions of a burgeoning quantity of medical diagnostic tasks where AI and ML are proving more accurate than humans. AI is also untiring and no matter how many thousands of scans it goes through, its accuracy will not decrease (providing it is well maintained), unlike with humans.

Allows Less Qualified Medical Staff To Make Decisions

Hierarchies of qualification in the medical industry limit the ability for some humans to make decisions. AI can make the decision on behalf of doctors so nurses and assistants do not always need to wait for doctor’s orders to carry out basic medical tasks.

Predicting Outcomes Better For Patients

Healthcare ML algorithms generate extremely detailed statistics, and can simulate a disease from start to finish with remarkable accuracy.

The quality of your machine learning prediction is affected by several factors:

• The machine learning algorithm.
• Data quality.
• The volume and variety of the data.
• The unique features which are selected to help predict the target variable.

Using a machine learning model can decrease the time needed to study and diagnose a patient’s condition, leading to time-efficiency and allowing experts to act without delay.

---

Databasing And Crowdsourcing

Medical information is highly sensitive, detailed and complex. Working with it can be notoriously hard. AI provides an efficient and secure means to store medical data and allow it to be accessed securely. Databases can be maintained and enhanced by healthcare facilities across the world to build very strong centres of research and knowledge. 

Some key examples of publicly available datasets that can be used to train ML algorithms include:

• INbreast and BreakHis for breast cancer testing and Digital Database for Screening Mammography (DDSM) for breast cancer detection.
• MITOSTAPIA for mitosis detection.
• Japanese Society of Radiological Technology (JSRT) database.
• Dermoscopic Image Segmentation (DermIS) database.
• The Lung Image Database Consortium (LIDC) database.
• Image Database Resource Initiative (IDRI) database.
• Multimodal Brain Tumor Segmentation challenge (BraTS) for brain cancer identification.
• Danish Lung Cancer Screening Trial (DLCST) for lung cancer classification.

---

GitHub Repository – Health Open Dataset

Health Open Dataset collects data from Dutch healthcare to offer insights, analysis, and other vital information. It uses web crawling and big query to collect public data and has a particularly rich source of data on coronavirus. 

---

GitHub Repository – COVID19

Designed to provide structured data on coronavirus in easily utilisable format. Designed to crowdsource all forms of COVID data for easy access by data scientists and practitioners. 

---

Machine Learning in the Healthcare Industry: Uses Cases

Diagnostics and Identification Of Diseases

ML algorithms can effectively diagnose disease and health conditions using computer vision and other data analysis techniques. Predictive models using neural networks or Bayesian networks can take into accounts thousands of biological variables.

For example, for cancer diagnosis, ML algorithms can assess tissues alongside genomic information and other markers (e.g. cytokines for inflammation) to create a detailed predictive model of prognosis and potential outcomes. ML can be used in the diagnosis and assessment of other conditions too, for example. ML has been used to predict the onset of mental health crises and identify what drugs and/or clinical interventions are appropriate to change outcomes.  

Cancer Diagnosis using Deep Learning – a Bibliographic Review

This medical review on cancer diagnosis provides Python codes for testing diagnostic techniques from a range of cancer data including size, shape, form and density. It uses many of the publicly available databases displayed above. The purpose of this review is to teach data engineers and scientists the basics of how ML algorithms using deep learning can be trained using public datasets on a range of cancers. 

GitHub Repository – Sytora

This is a symptom-disease classification app available across many languages. It works simply by running an algorithm to provide a list of diseases based on one or more input symptoms. 

GitHub Repository – ADS Screening

A data analysis toolset designed to discover patterns in autism datasets. This dataset contains 20 features of autism spectrum disorder that can be used for further analysis into the underlying cause, etc. It states that autism data is limited to genealogical factors and that more novel features are required for novel analysis.

---

Imaging with Computer Vision

A key area of medical diagnostics is imaging. Imaging also poses a unique problem that AI can help solve, as it is particularly vulnerable to human error. Radiologists are considered to possess some of the greatest skills of all medical practitioners when it comes to spotting minuscule abnormalities on X-rays and MRI scans.

Discovering these abnormalities quickly can be a matter of life and death. Despite their skill, radiologists are still subject to human error and these become vastly magnified with tiredness or distraction. ML algorithms using computer vision are changing the fields of radiology and medical imaging and can accurately read and identify a vast quantity of major and minor abnormalities with great accuracy.

---

Google DeepMind and Occipital Coherence Tomography

Google DeepMind can use occipital coherence tomography (OCT) scans, which are complex 3D images of the eye, to diagnose more than 50 eye diseases with remarkable accuracy. The issue with OCT scans is that they’re relatively easy to generate from eye imaging but very hard to read by medical professionals. Training ophthalmologists and opticians to read these sorts of scans is tricky but DeepMind can do so with remarkable speed and accuracy.

---

Drugs & Medication

Machine learning is proving incredibly useful in the research and manufacture of drugs. It can analyse datasets and simulate accurately how a drug could work as well as any side effects and long term impacts. Able to sift and analyse vast quantities of information collected during clinical trials, ML can cut data processing and analysis time by 70%. ML algorithms can analyse drug data profiles prior to trials or manufacture.

Want to improve your data skills?

See the best data engineering & data science books

Shop Now

Personalised Treatment

Machine learning has already been used to measure how effective drugs might be in treating diseases and conditions, including automatically appropriating the right dosage for the patient. This yields exciting potential for an array of highly personalised treatments. By incorporating the data and utilizing various problem-solving algorithms, ML offers suggestions on next-gen medication, e.g. tailoring precise treatment to the unique genomic profile of a tumour. 

GitHub Repository – Hackillinois

A PMC that can pull up patient history and determine automatically whether newly prescribed medication may cause adverse reactions. 

GitHub Repository – Diabetic Readmission Prediction

Useful for personalised treatment, this predictive platform uses ML to predict and determine the chance of diabetics being readmitted for treatment. Bases predictions on age, gender, medication, hospital stay duration, test results, exam results and vital signs. Uses 10 years of records from over 130 hospitals and 50 unique features.

 

---

Behavioural and Cognitive Science 

ML in psychology, psychometry and cognitive science has a rather uneasy relationship off the back of Cambridge Analytica and their manipulative tactics involving advertising and social media for political gain. However, in healthcare, ML yields exciting perspectives for neuroscientists, cognitive scientists and psychologists when simulating an array of neural processes to glean valuable insights into the way the brain works. This is potentially very useful in understanding and treating addiction and diagnosis of a range of cognitive conditions and diseases including Alzheimer’s. 

Indeed, cognitive science and ML engineering are intrinsically linked when developing increasingly human-like systems. The ways in which human learning builds very detailed rich and detailed models of the world using strong and accurate generalisations, even despite the world being packed full of noisy, sparse or ambiguous data, provides an incredibly fertile backdrop for engineering ML to replicate some of our own cognitive abilities.

---

Automated Surgery and Medical Procedures

As we’ve seen, ML already plays a role in diagnosis, but what about treatment via actual surgery? A robot in Massachusetts has already ‘bested’ highly trained human surgeons in performing complex heart valve transplants in pigs. The Versius robot system due to roll out to more healthcare facilities in 2020 enables ‘minimal access surgery’, where incredibly small insertions are made to carry out what would otherwise be very invasive surgery when carried out by humans.  

Many of these robots use ML and are trained on vast simulative datasets of real-world operations. Combined with advanced computer vision and machine precision, they are successful in many types of surgery ranging from hip replacements to brain surgery.

---

Conclusion

AI and ML in healthcare create value through its ability to work with large sets of complex data with precision. In clinical settings, not being prone to human error is a massive advantage.

Clinical settings are not always ideally suited to judgement calls and though humans are remarkably good at decisionmaking in general, AI and ML is not vulnerable, say, when a doctor has been on their feet for 16 hours and needs to make a life or death call, or when a radiographer is sifting through their 1000th MRI scan or X-ray.

In these cases, robots using the power of ML could vastly improve healthcare outcomes by appropriating accurate diagnosis and treatment much quicker than human medical teams. This even extends to surgery where robots are able to support or even replace human surgeons, potentially outperforming them across even the most difficult operations. 

Finally, ML’s great efficiency in working with data means it’s highly useful in assessing drug interactions, sifting through results in clinical trials or performing a vast systematic analysis of worldwide health data. It can also scrape and structure data that is already present in public datasets to make it more readily available for training further ML algorithms or for general use across healthcare sectors.

---

FAQs