Computers Catching Cancer

If there’s one thing on which all pathologists agree, it’s that their workloads are becoming increasingly untenable. It is a discussion of increasing importance (see “All in a Day’s Work") – and with a growing patient population and a shortage of trainees entering the profession, solutions are difficult to find. Enter a promising pathology assistant: the computer.

With the rise of digital pathology, fewer and fewer pathologists are strangers to computer-aided diagnosis, but a new deep-learning computer network developed by researchers at Case Western Reserve University significantly ups the ante. The network demonstrated 100 percent accuracy in detecting and delineating invasive breast cancers in whole biopsy slides, and made the same determination in each individual pixel 97 percent of the time – exceeding the accuracy and consistency of the four pathologists against which it was tested (1).

So is it time to replace the human brain at the microscope with a digital one? Not just yet. “The network was really good at identifying the cancers, but it will take time to get up to 20 years of practice and training of a pathologist to identify complex cases and mimics, such as adenosis,” said Anant Madabhushi (2), study co-author and Director of the university’s Center of Computational Imaging and Personalized Diagnostics. Instead, he proposes that the network could triage cases for review by pathologists, saving time and allowing them to focus their attentions on the samples – and the patients – who need it most. “If the network can tell which patients have cancer and which do not, this technology can serve as triage for the pathologist, freeing their time to concentrate on the cancer patients.” And best of all, the software can be set to run independently while pathologists work (or sleep), alleviating the intensifying burden on pathology department staff.