by
Lauren Dubinsky, Senior Reporter | August 19, 2016
Field is entering into the 21st century
Trained computers are able to predict lung cancer type and severity more accurately than pathologists, according to a new study by Stanford University School of Medicine. This finding may be what brings the pathology field into the 21st century.
"Pathology as it is practiced now is very subjective," Michael Snyder, professor and chair of genetics at the university, said in a statement. "Two highly skilled pathologists assessing the same slide will agree only about 60 percent of the time."
Pathologists traditionally assess the severity or grade of cancer by using a light microscope to examine thin cross-sections of tumor tissue on glass slides. The grade is higher depending on how abnormal the tissue looks in terms of cell size and shape.
The cancer's grade and stage can often be used to predict how the patient will fare and help the physician determine how to treat it. But that approach doesn’t work well for lung cancer because the cell subtypes can be difficult to differentiate.
In addition, the stage and grade of the cancer doesn’t always coincide with the prognosis. For example, 50 percent of stage 1 adenocarcinoma patients die within five years of diagnosis but about 15 percent survive over 10 years.
The researchers trained the computer using 2,186 images of adenocarcinoma and squamous cell carcinoma from a national database called the Cancer Genome Atlas. The database also contained information about the grade and stage of each cancer and how long the patient lived after diagnosis.
The computer was instructed to identify cancer-specific characteristics that can be detected with the human eye. The computer can spot 10,000 individual traits, including cell shape and texture, compared to the several hundred that the pathologist assesses.
The researchers focused on a group of cellular characteristics that the computer software found that could be used to distinguish tumor cells from the surrounding tissue, identify cancer subtype and predict how long each patient will survive post diagnosis.
They then used a data set of 294 lung cancer patients from the Stanford Tissue Microarray Database to validate the software’s ability to accurately distinguish short-term survivors from those who lived significantly longer.
These previously unknown physical characteristics that can predict cancer severity and length of survival may also provide a better understanding of the molecular processes of cancer initiation and progression. Snyder is hopeful that the software will complement the emerging fields of cancer genomics, transcriptomics and proteomics.
Back to HCB News
