E-News Exclusive
AI May Improve Diabetes Diagnosis
Using a fully automated AI deep learning model, researchers were able to identify early signs of type 2 diabetes on abdominal CT scans, according to a recent study published in the journal Radiology. Type 2 diabetes affects approximately 13% of all adults in the United States, and an additional 34.5% of adults meet the criteria for prediabetes. Due to the slow onset of symptoms, it is important to diagnose the disease in its early stages. Some cases of prediabetes can last up to eight years, and an earlier diagnosis will allow patients to make lifestyle changes to alter the progression of the disease.
Abdominal CT imaging can be a promising tool to diagnose type 2 diabetes. CT imaging is already widely used in clinical practices, and it can provide a significant amount of information about the pancreas. Previous studies have shown that patients with diabetes tend to accumulate more visceral fat and fat within the pancreas than nondiabetic patients. However, not much work has been done to study the liver, muscles, and blood vessels around the pancreas, says study cosenior author Ronald M. Summers, MD, PhD, a senior investigator and a staff radiologist at the National Institutes of Health Clinical Center in Bethesda, Maryland.
"The analysis of both pancreatic and extrapancreatic features is a novel approach and has not been shown in previous work to our knowledge," says first author Hima Tallam, BSE, an MD/PhD student.
The manual analysis of low-dose noncontrast pancreatic CT images by a radiologist or trained specialist is a time-intensive and difficult process. To address these clinical challenges, there is a need for the improvement of automated image analysis of the pancreas, the authors say.
For this retrospective study, Summers and colleagues, in close collaboration with cosenior author Perry J. Pickhardt, MD, a professor of radiology at the University of Wisconsin School of Medicine and Public Health in Madison, used a data set of patients who had undergone routine colorectal cancer screening with CT at the University of Wisconsin Hospital and Clinics. Of the 8,992 patients who had been screened between 2004 and 2016, 572 had been diagnosed with type 2 diabetes and 1,880 with dysglycemia, a term that refers to blood sugar levels that go too low or too high. There was no overlap between diabetes and dysglycemic diagnosis.
To build the deep learning model, the researchers used a total of 471 images obtained from a variety of data sets, including the Medical Data Decathlon, The Cancer Imaging Archive, and the Beyond Cranial Vault challenge. The 471 images were then divided into three subsets: 424 for training, eight for validation, and 39 for test sets. Researchers also included data from four rounds of active learning.
The deep learning model displayed excellent results, demonstrating virtually no difference compared with manual analysis. In addition to the various pancreatic features, the model also analyzed the visceral fat, density, and volumes of the surrounding abdominal muscles and organs. The results showed that patients with diabetes had lower pancreas density and higher visceral fat amounts than patients without diabetes.
"We found that diabetes was associated with the amount of fat within the pancreas and inside the patients' abdomens," Summers says. "The more fat in those two locations, the more likely the patients were to have diabetes for a longer period of time."
The best predictors of type 2 diabetes in the final model included intrapancreatic fat percentage, pancreas fractal dimension, plaque severity between the L1-L4 vertebra level, average liver CT attenuation, and body mass index. The deep learning model used these predictors to accurately discern patients with and without diabetes.
"This study is a step towards the wider use of automated methods to address clinical challenges," the authors say. "It may also inform future work investigating the reason for pancreatic changes that occur in patients with diabetes."
— Source: RSNA
AI Shows Potential in Breast Cancer Screening Programs
A recent study in Radiology shows that AI is a promising tool for breast cancer detection in screening mammography programs. Mammograms acquired through population-based breast cancer screening programs produce a significant workload for radiologists. AI has been proposed as an automated second reader for mammograms that could help reduce this workload. The technology has shown encouraging results for cancer detection, but evidence related to its use in real screening settings is limited.
In the study, the largest of its kind to date, Norwegian researchers led by Solveig Hofvind, PhD, from the Section for Breast Cancer Screening, Cancer Registry of Norway in Oslo, compared the performance of a commercially available AI system with routine independent double reading as performed in a population-based screening program. The study drew from almost 123,000 examinations performed on more than 47,000 women at four facilities in BreastScreen Norway, the nation’s population-based screening program.
The data set included 752 cancers detected at screening and 205 interval cancers—cancers detected between screening rounds. The AI system predicted the risk of cancer on a scale from 1 to 10, with 1 representing the lowest risk and 10 the highest risk. A total of 87.6% (653 of 752) of screen-detected and 44.9% (92 of 205) of interval cancers had the highest AI score of 10.
The researchers created three thresholds to assess the performance of the AI system as a decision-making tool. Using a threshold that mirrors the average individual radiologist rate of positive interpretation, the proportion of screen-detected cancers not selected by the AI system was less than 20%. While the AI system performed well, the study’s reliance on retrospective data means that more research is needed.
“In our study, we assumed that all cancer cases selected by the AI system were detected,” Hofvind says. “This might not be true in a real screening setting. However, given that assumption, AI will probably be of great value in interpretation of screening mammograms in the future.”
The results showed favorable histopathologic characteristics associated with a better prognosis for screening-detected cancers with low vs high AI scores. Opposite results were observed for interval cancers. This may indicate that interval cancers with low AI scores are true interval cancers not visible on the screening mammograms.
The high percentage of true negative examinations classified with a low AI score has the potential to substantially reduce the interpretive volume, while allowing only a small proportion of cancers to go undetected. By using AI as one of the two readers in a double reading setting, the radiologist could still identify these cancers, the researchers say.
“Based on our results, we expect AI to be of great value in the interpretation of screening mammograms in the future,” Hofvind says. “We expect the greatest potential to be in reducing the reading volume by selecting negative examinations.”
Although more study is needed before clinical implementation of AI in breast cancer screening, the results of the study help establish a basis for future research, including prospective studies, Hofvind says.
“We are looking forward to testing out different scenarios for AI using retrospective data and then running a prospective trial,” she says.
— Source: RSNA
