Page 16 - Layout 1

MEDICAL YEAR
IN REVIEW

Artificial intelligence (AI) and medical
imaging are merging to create more precise

diagnostic tools for pathologists and better
treatment options for cancer patients. Using

a type of AI known as machine learning, we
program computer systems to “learn” or
DETECTING AND simulate human processes, such as visual

ANALYZING perception and decision making. The

systems form a neural network of
BREAST CANCER algorithms that calculate the best solutions

for a wide range of problems. Recently, our
CELLS team of engineers at Southwest Research

Institute (SwRI) used AI to enhance cancer
By Hakima Ibaroudene diagnostics. In collaboration with UT Health

San Antonio pathologists, we trained a
computer algorithm to quickly and

accurately detect and analyze breast cancer
tumor cells. The team envisions developing

this algorithm further, eventually using it to
detect other forms of cancer and to collect

valuable, life-saving data from cancer cells,
such as DNA structure and mutation

analysis information.

The Cancer Cellularity Challenge tumors. Our SwRI engineers sought the ex- the other to test it. Once our team was sat-
The journey to developing a cancer-de- pertise of the UT Health pathologists who isfied with the algorithm, we analyzed the
tecting algorithm began with an interna- condensed years of training into a short images from breast cancer patients and as-
tional competition, the BreastPathQ: course on pathology and imaging. Armed signed a score based on the number of in-
Cancer Cellularity Challenge conducted by with new knowledge on the appearance vasive cancer cells in each image. The
the American Association of Physicists in and structure of breast cancer tumor cells, algorithm sorted through the images and
Medicine, the National Cancer Institute our team set out to train the algorithm to matched the findings of human pathologists
and SPIE, the international society for op- analyze cell images and look for defining at the highest rate, making it the top-per-
tics and photonics. The challenge pre- characteristics that distinguish cancerous forming algorithm out of 100 competing
sented the task of determining cancer cells from normal ones. submissions.
cellularity from pathology hematoxylin and Challenge organizers provided two collec- In January 2019, we learned we won the
eosin (H&E) slide patches of breast cancer tions of images: one to train the algorithm, international competition. Our success on

16 San Antonio Medicine • December 2019