A diagnostic report can change the course of a patient’s treatment and thus the accuracy and timing of diagnosis of a disease is critical in a patient’s clinical management and cure. Here’s a look at how artificial intelligence has the potential to work in tandem with doctors, complementing their expertise and experience.
Over the years, medicine has become more and more evidence-based, making diagnostics a key aspect of any clinical treatment. The pathological diagnosis is usually considered the gold standard for patient management. However, examining hundreds of tissue slides is a complex process, requiring years of experience and learning to report malignancy or traces of any abnormalities. Further, the low rate of concordant results on the same slides by different pathologists only aggravate the problem further.
While an accurate diagnosis is invaluable, a diagnostic error can result in delayed, inappropriate or wrong treatment. Based on the postmortem examination research spanning decades reveals that “diagnostic errors contribute to approximately 10% of patient deaths, and medical record reviews suggest that they account for 6-17% of adverse events in hospitals.”
Given the humongous amount of information that must be reviewed to make a decision, the lack of consensus and errors in reporting cannot be ruled out. To streamline the diagnostic variability within limited time-frame, advanced technologies such as machine learning are being applied to review pathological slides to complement the expertise of pathologists, reduce workload pressures and enhance accuracy in reporting.
In mid-2017, the research team led by Case Western Reserve University used its deep-learning computer network achieving 100% accurate results in determining the presence of invasive forms of breast cancer in whole biopsy slides.
The study said, “Network training took about two weeks and identifying the presence and exact location of cancer in the 200 slides took about 20 to 25 minutes each.”
Around the same time, Philips IntelliSite Pathology Solution (PIPS) by Philips (PHG) became the first digital pathology solution to receive FDA clearance for primary diagnostic use in the U.S. pathology labs. The technology assists pathologists to review and interpret digital images of surgical pathology slides prepared from biopsied tissues.
In 2017, Google (GOOG, GOOGL) described its deep learning–based approach (LYmph Node Assistant, or LYNA) to improve diagnostic accuracy and achieved better results for cancer detection. In 2018, it published two papers on the subject, where it stated that, “LYNA was able to correctly distinguish a slide with metastatic cancer from a slide without cancer 99% of the time (for the datasets used).” Google is also working on an augmented microscope that enables real-time image analysis and presentation of the results of machine learning algorithms directly into the field of view.
In addition to the research on use of algorithms for assisting pathologist in detecting breast cancer, Google is applying computer-aided diagnostic screening for diabetic retinopathy. It is estimated that around 28.5% of U.S. diabetics have diabetic retinopathy, which can lead to blindness.
Beyond pathology, various aspects of medical imaging are leveraging AI. Last year, the FDA permitted the marketing of artificial intelligence algorithm for aiding providers in detecting wrist fractures. The OsteoDetect software uses an artificial intelligence algorithm to analyze two-dimensional X-ray images for signs of distal radius fracture, a common type of wrist fracture.
Another interesting example is from a project dubbed InnerEye by Microsoft Research. Its purpose is utilizing computer vision and machine learning to build tools for automatically analyzing 3D radiology images. The project’s main focus is in the treatment of tumors and monitoring the progression of cancer in temporal studies.
Nuance (NUAN) offers a marketplace for diagnostic imaging, providing developers immediate access to 70% of all radiologists across 5,500 connected healthcare facilities.
NVIDIA (NVDA) has achieved success in predicting Alzheimer’s disease from resting-state functional MRI (rs-fMRI) data by using NVIDIA DIGITS to train a Convolutional Neural Network model. Alzheimer’s is the sixth-leading cause of death in the U.S.; the direct costs of caring for those with Alzheimer's will total an estimated $277 billion in the U.S., a figure projected to inflate to $1.1 trillion by 2050 (in 2018 dollars), according to the Alzheimer's Association.
A study that used NVIDIA’s TITAN Xp GPUs with CUDA 9 and cuDNN 6 reports that deep learning can help predict Alzheimer’s disease six years before an actual diagnosis.
“If we can detect it earlier, that’s an opportunity for investigators to potentially find better ways to slow down or even halt the disease process,” says study co-author Jae Ho Sohn, MD, from the Radiology & Biomedical Imaging Department at the University of California in San Francisco (UCSF).
Overall, artificial intelligence (which encompasses deep learning, computer vision, robotics, collaborative systems, machine learning and natural learning process among other things) has the potential to speed up diagnoses, improve accuracy and efficiency, all of which would positively impact patient outcomes through correct follow up treatment.
The author has no position in any stocks mentioned. Investors should consider the above information not as a de facto recommendation, but as an idea for further consideration.