With over 12 million Americans affected by medical diagnosis errors every year, healthcare professionals turn to technology to reduce the percentage of misdiagnosed patients amid the aggravating workforce shortages.
One of the most potent technologies in medical diagnosis is machine learning (ML) — a subset of artificial intelligence (AI) that relies on smart algorithms trained on labeled or unlabeled data.
Depending on the machine learning type, these algorithms can process both structured data, which exists in the form of neatly organized tables in databases, and unstructured data, such as medical images, non-editable documents, and sensor readings.
For example, machine learning played a critical role in Johnson & Johnson coronavirus vaccine development, while researchers from South Korea leveraged ML models to predict mortality rates among COVID-19 patients.
In this article, a team of innovation analysts from Symfa a custom software development company focussing on business transformation, will dive into ML applications in healthcare — specifically, medical diagnosis (Dx).
Read on to find out whether algorithms will soon replace your GP!
Machine Learning in Healthcare: Top 5 Use Cases in Medical Diagnosis
- Machine learning in oncology. ML applications in oncology span medical image and patient data analysis. Speaking of medical images, custom-trained models can help physicians detect tumors in various imaging modalities, such as mammograms and MRI, CT, and PET scans. Following neoplasm detection, ML software development can further help measure the tumor boundaries and nature (malignant or benign). Oncologists also use ML to predict an individual’s susceptibility to certain types of cancers by analyzing their genomic data — or identify patterns related to cancer occurrence, progression, and outcomes in a particular patient group by conducting ML-driven EHR data analysis. For certain types of cancers, such as breast cancer and melanoma, machine learning patient diagnosis tools already perform on a par with human oncologists or even demonstrate superior results. However, the maximum effect is achieved when physicians use AI-based cancer detection tools for preliminary image analysis. In this case, oncologists can improve diagnostic accuracy by up to 11.5% while significantly reducing the workload.
- Machine learning in pathology. Machine learning — particularly, its deep learning branch that involves feeding unstructured data to algorithms — is the go-to technology for analyzing high-resolution tissue images in digital pathology. For example, ML models can be trained to identify patterns associated with inflammatory diseases, infections, and, as we mentioned earlier, cancers. Algorithms can also assess a tumor’s malignancy and aggressiveness. Additionally, ML can aid healthcare professionals in predicting gene mutations or other molecular anomalies based on tissue morphology and provide consistent, quantifiable metrics, such as the percentage of tumor cells or the extent of necrosis in tissue samples. Some early practical examples of machine learning in pathology include the Google Research team’s partnership with the Medical University of Graz and Biobank Graz. The experiment involved training a deep learning model to predict treatment outcomes in patients with colorectal cancer based on pathology images. Despite proving a success, the study highlighted the importance of collaboration between ML engineers and expert pathologists at all stages of the model training and adjustment process.
- Machine learning in ophthalmology. ML solutions demonstrate excellent performance in the diagnosis of various eye disorders, including diabetic retinopathy, cataracts, glaucoma, and age-related macular degeneration (AMD). For this, software engineers train custom models on retinal photographs, optical coherence tomography (OCT) scans, and fundus images. Outstanding results have been so far achieved in diabetic retinopathy diagnosis. One example of ML-assisted image analysis comes from Google engineers, whose AI Eye Doctor solution can spot early signs of the disease and tell whether diabetes can affect a patient’s eyesight in the longer run with a 98.6% accuracy, rivaling human ophthalmologists.
- Machine learning in cardiology. Feeding off electrocardiography (EKG) data, as well as CT and MRI scans, machine learning models can detect and classify arrhythmias, diagnose congenital heart diseases (CHD), valve disorders, and cardiomyopathies, and predict the likelihood of myocardial infarction and heart failures in high-risk patients. Back in 2019, an international team of researchers from Brazil’s National Institute of Space Research and Max Planck Institute for Brain Research (Germany) created an ML-based EKG arrhythmia classifier with a positive predictive value of over 75%. In another example, US researchers trained AI algorithms to identify patients with heart disease who were at a higher risk of dying within 12 months. Such technologies have the potential to enhance diagnostic procedures in cardiology and help physicians take a preventive approach to heart disease treatment.
- Machine learning in virology. Healthcare ML solutions have long proven effective in studying viral genomes, spotting mutations, predicting outbreaks of viral diseases, and modeling their transmission routes. When it comes to medical diagnosis, deep learning networks can help radiologists identify signs of viral infections in X-ray and CT images. The intelligent algorithms can also analyze patient samples to determine the presence of a particular virus based on its unique genetic or proteomic markers. And during the COVID-19 pandemic, scientists leveraged machine learning to diagnose the virus in patients with flu-like symptoms based on their cough sounds.
These and many other ML applications demonstrate the technology’s potential in medical diagnosis and hold immense promise for the healthcare sector.
The question remains: is the technology stable and mature enough to deliver consistently good results outside the lab walls, and will it eventually replace human doctors?
What’s in Store for Machine Learning in Medical Diagnosis
No matter how hard data scientists try, they’re unlikely to create machine learning models as complex and efficient as the human brain.
But ML solutions for medical diagnosis excel at one thing where humans fall short. We’re talking about processing large amounts of training data over a short period of time — and being able to make sound conclusions when faced with similar information.
In the coming years, we’ll see more examples of medical ML systems acting as a second pair of eyes to healthcare professionals. Such systems can be vital in preliminary diagnosis and, in some cases, may even pick up on signs of diseases that doctors miss. Otherwise, the future of medical diagnosis will remain human — at least for now.
