Opportunities for AI to assist Physicians are everywhere

Jun Wu

May 23, 2019·4 min read

Listen to your patient, he is telling you the diagnosis — William Osler

In the golden age of Artificial Intelligence, healthcare is the new frontier of research and development. Surgeons are routinely using robotic assists to operate with less invasiveness and more precision. Gene sequencing and gene editing aided by AI is transforming the way scientists obtain cures for diseases.

But, most notably, research is underway to allow AI to transform the way doctors diagnose patients.

Picture this:

You have symptoms of a cold. But, your fever hasn’t subsided in days. Your breathing is constricted. You drive yourself to the hospital.

In the hospital, your doctor asks you about your symptoms and records them into the doctor’s iPad.

Within a few seconds, an AI diagnosis system pulls up all your medical history, puts your symptoms into the context of your medical history, then gives the doctor a recommendation of your diagnosis.

The doctor looks over the diagnosis and compares it with his/her personal evaluation.

Then, the doctor discusses this diagnosis with you.

There’s no waiting for hours for a diagnosis.

This is the future of medical diagnosis — an AI Diagnostic System to assist doctors in diagnosing all kinds of diseases.

This future is pretty close. In some areas of healthcare, the future is already here.

Kang Zhang at the University of California in San Diego and his colleagues trained an AI on medical records from 1.3 million patient visits at a major medical center in Guangzhou, China. The AI is able to diagnose glandular fever, roseola, influenza, chicken pos and hand-foot-mouth disease with 90 and 97 percent accuracy.

In a study published in Nature Medicine, using Google’s AI system, sourcing data 42,000 patient CT scans, the AI system outperformed 6 radiologists in diagnosing lung cancer. It improved the detection rate for lung cancer by 5%.

In April 2018, the FDA approved IDx-DR, a tool using AI to detect diabetic retinopathy by studying the images of the back of the eye. In the clinical study that led to the approval, the system detected diabetic retinopathy correctly 89.5% of the time. It’s the first tool of its kind that can be used for diagnosis without a doctor’s intervention.

With speedy innovation, there are also challenges.

Technology Behind AI for Diagnosis

These AI Systems for diagnosis uses deep learning techniques to arrive at a diagnosis. These systems source image data and symptoms data from healthcare facilities to train on. Then, the systems apply the techniques to arrive at the diagnosis. The problem with deep learning techniques is transparency. Although the inputs and the outputs to the AI systems are transparent, the way that the AI systems arrive at the diagnosis decision is unclear. The system is also dependent on the quality of the data to ensure accuracy.

Challenges of Sourcing Data

In the U.S., healthcare data is located in various healthcare facilities, at insurance companies, and inside government agencies. Sourcing all of the data into a data cloud for AI Systems to use is a huge task. Google is currently embarked on building this exact infrustructure for medical technology companies and healthcare facilities to use to train their AI Systems. It is taking significant amount of investment for Google to step through all the regulations and privacy rules to provide this infrastructure.

Challenges aside, the future looks bright for AI in medical diagnosis.

Medical Imaging

One of the newest areas for using AI Systems to both automate the workflow for efficiency as well as assisting in diagnosis is in Medical Imaging.

Medical images, such as X-ray, ultrasound, CT or MRI scan, can be used to diagnose a variety of diseases. Radiologists currently review such scans for diagnosis. Often, depending on the quality of images, radiologists can make errors due to the limitations of the human eye or the lack of experience in a specific disease area. Review of the images often takes time.

In the ER, when time is critical, a review of the images in a timely fashion can lead to life-saving results.

AI Systems can apply its “advanced vision” to scan medical images for abnormalities. Using training data, the AI Systems can spot these abnormalities and bring them up for review by the radiologist. Alternatively, autonomous AI Systems can make the diagnosis efficiently for diseases that have clear cut causes that are easy to find for the AI System.

The repetitive work of reviewing the scans to look for abnormalities is automated. Once the abnormalities are identified, the radiologists can then apply his or her experience to make the actual diagnosis.

The improved efficiency in the Radiology workflow results in a more accurate diagnosis as well as a more timely diagnosis.

At the forefront of innovation is Nvidia Clara, an AI-powered clinical imaging platform that processes medical imaging to improve time to diagnosis and provide early warnings to doctors particularly in the ER departments of hospitals.

Also notable are: Subtle Medical, a silicon valley startup that develops tools to improve the productivity of the radiology workflow; Imfusion, another startup that’s building AI tools to turn 2D ultrasound into 3D images.

Conclusion

As AI innovations progress in healthcare, it’s critical for both medical technology companies, data providers (Google), hospitals, governments, and insurance companies to work together to foster an atmosphere of innovation. In this atmosphere of innovation, better opportunities for automation can be identified, proper data can be sourced, thus lead to better accuracy in both medical diagnosis and better efficiency in the healthcare workflow.