Notwithstanding the serious concerns raised by a recent fatal accident involving a Tesla car running on autopilot, self-driving cars are probably here to stay—but that doesn’t mean humans won’t still be driving. The same goes for fully automated abdominal CT image interpretation. It too is likely here to stay—but that doesn’t mean radiologists won’t still be reading.
Ronald Summers, MD, PhD, of the NIH’s laboratory for imaging biomarkers and computer-aided diagnosis, explores the state of the latter scenario in the July edition of the American Journal of Roentgenology.
Automated analysis of abdominal CT “has advanced markedly over just the last few years,” Summers writes. “Fully automated assessment of organs, lymph nodes, adipose tissue, muscle, bowel, spine, and tumors are some examples where tremendous progress has been made. Computer-aided detection of lesions has also improved dramatically.”
Summers reviews the technology as it has already been applied to each of those anatomic structures and various types of lesions, and he notes that there are two approaches to developing software capable of automatically and accurately analyzing radiologic images.
In one, the software developer concentrates on “handcrafting” ways to distinguish particular pathologies from normal tissue.
In the other, the developer uses a machine-learning algorithm to teach the software to tell disease from non-disease by being trained on labeled cases, without the need for handcrafted features.
This second approach is “made feasible by recent advances in computer science known colloquially as deep learning,” Summers notes, adding that it is “increasingly being used because it markedly increases the efficiency of image analysis development.”
“To perform fully automated abdominal CT image interpretation at the level of a trained radiologist, the computer must assess all the organs and detect all the abnormalities present in the images,” Summers writes. “Although this is a seemingly daunting task for the software developer, the numbers of organs and potential abnormalities are finite and can be addressed methodically.”
Summers next summarizes related areas of activity with which radiologists will soon need to reckon. Among the areas he spotlights:
- Deep machine learning. Deep learning, which is a shorthand term for convolutional neural networks, has “led to dramatic improvements in the performance of three different body CT CAD systems,” Summers writes. Researchers “improved existing CAD systems for sclerotic spine metastases, lymphadenopathy and colonic polyps. Sensitivities improved from 57 percent to 70 percent, from 43 percent to 77 percent, and from 58 percent to 75 percent, respectively.”
- Big data and automated radiology reporting. CT scans can be analyzed alongside their accompanying reports, which can facilitate the efficient linking of report findings with image findings. “In a study of 216,000 radiology key images identified during routine clinical interpretation of scans from 62,000 unique patients, including many from abdominal CT scans,” Summers writes, “the rate of predicted disease-related words matching the actual words in the report sentences was 56 percent.”
- New applications to improve patient care. Here Summers mentions the potential of fully automated image interpretation in underserved populations to meet global health needs. He also notes the promise of the technology to run “in the background,” allowing clinicians to more efficiently triage patients, reduce errors, conduct studies and coordinate mega datasets combining clinical, genomic and imaging information.
“As radiology practices consolidate into larger hospital-led groups, it will be more feasible to implement such systems,” Summers states. “Such big data analyses could uncover previously unrecognized associations among imaging findings, drug treatments and other data in the clinical record. These applications promise improvements in patient care.”
Summers exhorts radiologists to keep their skills up to speed as technological developments continue to accrue.
“The automated report could improve reading efficiency, but radiologists will need to be vigilant to avoid placing too much trust in the computer,” he writes.
“Advances in abdominal CT automated image interpretation are occurring at a rapid pace,” Summers concludes. “In the not too distant future, these advances may enable fully automated image interpretation. Similar advances may occur in other body regions and with other imaging modalities.
“Risks and benefits are difficult to foresee but may include increased pressures for commoditization, better reading efficiency, fewer interpretive errors and a more quantitative radiology report. The primary focus must ultimately be on improved patient care.”