Pathologists have a great responsibility when it comes to cancer diagnosis. Based on the data of the same patient, doctors disagree on diagnosis up to 48% of the time. Now, Google’s deep learning AI system can detect cancer faster, and with higher accuracy than human professionals.
Early and accurate cancer diagnosis is vital for subsequent treatment especially for virulent types of cancer like mesothelioma. Pathologists are limited to the examination of scans which show possible tumors, a lengthy and complicated procedure given the huge amount of data they have to sift through.
As automation grows, it is natural to worry about human job security. Yet, it is much more likely that these systems will enable us to expand our capabilities and assist us with grueling, difficult tasks.#Google deep learning AI diagnoses cancer more accurately than humans.Click To Tweet
The Human eye and Cancer Detection
One patient’s file contains many slides of tissue samples, each more than 10 gigapixels when magnified at 40X, which means the need to review thousands of images, pixel by pixel.
With this level of detail, reviewing the slides requires extreme precision and a trained eye to detect abnormalities and the occurrence of metastases. A time-sensitive and error-prone job, since even an experienced staff may make errors. This leads to diagnostic variability and impacts the accuracy of human professionals in detecting cancer, estimated at about 73.2%. Even then, doctors agree on the same diagnosis in only half of the cases based on the same clinical data.
Google’s AI: From Autonomous Cars to Cancer Detection
Via its GoogLeNet projects, Google has been working for several years to develop a sophisticated image recognition system with the intention to integrate it into its autonomous cars so they can distinguish everything in the roadway and the surrounding environment.
Alphabet, Google’s parent company, has taken this a step further in diversifying its activities to have a greater impact on human life. In a white paper, Detecting Cancer Metastases on Gigapixel Pathology Images, Google unveiled research using its deep learning AI for the diagnosis of cancer.
To test the system, Google’s experts used a data set of images courtesy of the Radboud University Medical Center. At first, the tumor probability heatmaps produced were noisy. Then, after customizing and training the model to examine the image at different magnifications, it exceeded the performance of human doctors. Google’s algorithm produced improved prediction heatmaps and its localization score reached 89%, higher than humans’ 73% and also, needless to say, the AI completed its diagnosis at a much faster rate.
Google’s AI, or any system for that matter, is not intended to completely replace pathologists, but to complement their work.
The system is able to determine the size of the tumor with great precision and detect certain inconsistencies that human professionals wouldn’t notice. It can differentiate between healthy and cancerous tissues and tell if a metastasis has occurred. AI provides high false-positive results and can, for example, submit suspected images to doctors for further examination.