考研阅读精选：如何准确诊断癌症？
『通过使用编制的计算机程序，医生可以准确无误的确诊癌症。』Indolent or aggressive?
如何准确诊断癌症？
Nov 8th, 2011 | from The Economist
http://images.koolearn.com/casupload/upload/fckeditorUpload/2011-11-28/image/61a53522bb994097a29972b67d541a58.jpg
LOOKING for needles in haystacks is boring. But computers do not getbored. Contracting out to machines the tedious business of assessing thedangerousness of cancer cells in histological microscope slides oughtthus to be an obvious thing to do. Cervical-cancer smear tests aside,however, such electronic intrusions into the pathology laboratory arelimited. Grading cancer cells into “indolent” and “aggressive”, andhazarding an opinion about whether they spell a treatable condition oran untreatable one, has remained the realm of the human expert. But notfor much longer, if Daphne Koller, a computer scientist at StanfordUniversity, and her colleagues have their way. They report in thisweek’s Science Translational Medicine that they have written a programwhich can distinguish between grades of breast-cancer cell—and can do soin a way that provides a more accurate prognosis than a humanpathologist can manage.
Previous attempts to build acomputerised pathologist of this sort involved the designers carefullyspecifying which characteristics of the samples being examined were mostimportant. For example, they would tell the computer to measure thethree traits human pathologists use to determine a tumour’s grade: thepercentage of its cells that are tubelike; the diversity of appearanceof the cell nuclei; and the proportion of cancer cells undergoingdivision. However, people are excellent at pattern recognition andskilled pathologists rely not just on these relatively-easy-to-describetraits, but also on less well defined characters that years ofexperience have taught them are significant too. Restrictingcomputerised pathologists to the well-characterised bits of the processtherefore inevitably results in worse performance than their humancounterparts show.
Dr Koller’s Computational Pathologist(C-Path), by contrast, lets the system work out for itself what the mostimportant features of a tumour are. She and her colleagues started bysetting down 6,642 characters the program might choose from when itassessed images of biopsies from breast-cancer patients, but did nottell it which to prefer. Some of the characters they offered wereinherent to the cancer cells. Others were features of the surrounding“stromal” cells, which are not, themselves, malignant, but act tosupport a tumour. And some were not features of individual cells at allbut, rather, measured relations between cells (for example, the averagedistance between cancer-cell nuclei) and the context cells foundthemselves in (for example, whether they occurred in large clusters orwere frequently interspersed with stroma).
The teaminitially trained and tested the program on 248 breast-cancer samplesfrom the Netherlands Cancer Institute. It was fed with images of slidesfrom these patients, and also information on how long each patient hadsurvived after the sample being examined had been taken. That done, theythen tested it on a second set of samples, this time from 286breast-cancer patients at Vancouver General Hospital. They found it wasable both to grade the slides and to predict—in a way human pathologistscould not—whether patient would survive for five years after treatment.
When Dr Koller looked at which 11 features were the mostrobust predictors of survival, she discovered that only eight werecharacteristic of the tumour cells themselves. The other three werestromal characters. The fact that three stromal features were on thelist suggests that the stroma influences whether or not a cancerprogresses and kills the patient. That is important information because,hitherto, pathologists have focused on the cancer cells themselves andignored the stroma. Thus C-Path seems not only to outperform humanpathologists, but also to have something to teach them about cancerbiology. (592 words)
文章地址：http://www.economist.com/blogs/babbage/2011/11/diagnosing-cancer
页:
[1]