The promise of artificial intelligence in diagnosing illness | Artificial intelligence

You visit your doctor for a routine checkup. She decides to do a fine needle aspiration for a seemingly insignificant lesion (“It’s probably nothing…”). But a few days later you get a call back. They’ve found some abnormal cells.

Your doctor recommends a specialist who performs a resection in his office, sending a tissue sample to a histology lab. The lab’s histologist preps the sample by taking sections, placing them in a “cassette,” then sending it overnight to another lab for processing. When the tissue blocks are returned the next day, a histologist cuts slides from the processed paraffin-embedded tissue and stains them with hematoxylin and eosin which differentially bind to cellular structures. A case number is created, whereupon the slides are placed on a plastic tray and delivered to the pathologist, as if serving hors d’oeuvres. The pathologist might walk the tray down the hall to another pathologist, who examines the specimens under a microscope, sometimes recommending additional interpretive stains. A diagnosis is made, and the pathology report is sent back to your doctor, who then creates a treatment plan. The entire process can take weeks, sometimes even months.

Patients and medical professionals have become accustomed to the drill: it’s been standard practice for decades. But advancements in artificial intelligence are about to disrupt disease diagnosis.

Diagnosis, accelerated

Imagine that a histologist can scan the slides, creating electronic images that can be accessed digitally by different physicians, labs, and technicians. Tissue samples can be stored in the cloud and shared with far-flung specialists who can collaborate from afar with care providers. Diagnoses can be corroborated through deep learning algorithms that recognize specific characteristics and behaviors in the tissue sample. The digital images become a historical record of the blood or tissue sample. The entire diagnostic process could go from weeks to hours, accelerating vital patient care.

“We’ve always had the discussion about whether pathology is art or science,” explains Dave Billiter, CEO and co-founder of Deep Lens, a digital pathology AI and cloud platform vendor. Billiter comes from more traditional healthcare, holding executive positions at Cardinal Health and The Research Institute at Nationwide Children’s Hospital. But he recognizes the of AI breaking new ground in medicine.

“In essence you’re relying on a trained eye to render a diagnosis of a biospecimen. But there are so many different types and subtypes of a disease. Even with highly trained medical professionals, you’re relying on an individual to understand them all. It’s an arduous process for a pathologist. Then you start adding all these different biomarkers, genomics, metabolomics… It’s a pathology information tsunami!”

As the need for more specialization increases, the number of pathologists is shrinking. According to the Medical Laboratory Observer, fewer medical school graduates are choosing pathology as a specialty, despite the increasing demand driven by an aging baby-boomer population and evolving genomic research.

“We see a decreasing pool of pathology experts,” confirms Saskia Boisot MD, a board-certified hematopathologist based in Orange County, California. “Part of it is that medical schools don’t feature careers in pathology as prominently as they do other specialties. Part of it is that the job market for pathologists is quite limited, the practice entirely dependent on affiliations with either hospitals or private reference laboratories, with little opportunity for autonomy.”

Humans, more than ever

Recognizing the opportunity, medical schools are beefing up their emerging technology curricula. Med students are now offered courses in technology infrastructure, deep learning, and data management alongside their biology classes, adding GPUs, robotics, and convolutional neural networks to their training.

Companies like Deep Lens and Philips package images in secure environments and can automatically place them into a pathologist workflow. AI algorithms can optimize routing to specialists and others in a dynamic, collaborative environment.

The benefits don’t stop at diagnosis. “Digitized results could move the needle for clinical trials,” says software investor and Deep Lens President and co-founder Simon Arkell. “Patient enrollment is difficult and time consuming. AI can identify a cancer, feed new models, and help researchers quickly qualify clinical trial participants. Drug companies and CROs [Contract Resource Organizations] could experience huge economies of scale.”

Then there’s the coming data deluge. According to Andrew Hessell, a biotech industry pioneer and now CEO of Humane Genomics, “It’s economically realistic for a billion human genomes to come online over the next decade or two.” It will be nearly impossible for pathologists to stay current with the emergence of new biomarkers without the help of machine learning and digitization.

“I think there’s a fear that, much in the same way a lot of radiology interpretation is being outsourced through digitization, pathologists could become progressively obsolete,” says Dr. Boisot. “Conventional anatomic pathology is undergoing a massive transformation, as many diagnoses are now being made based on molecular studies. While these studies currently only represent an adjunct to morphologic evaluation, there’s a very real chance that they could dwarf the need for a pathologist’s keen eye.”

Today’s problems have fomented over decades. But the pathology discipline isn’t going the way of the dinosaur. Billiter maintains that the pathology discipline will be central to personalizing medicine, with pathologists adopting new technologies drive faster, more accurate diagnostics. “This will be as much about humans helping AI as AI helping humans.” 

What better partnership could there be?

This article is published as part of the IDG Contributor Network. Want to Join?

You might also like More from author