Radiologists Tricked by AI (or Not), and Other Radiology News and Studies

Could a radiologist be tricked by AI?

As radiology sees more types of artificial intelligence (AI) algorithms come online to help with diagnosis, there has been a looming fear that such programs could be used as cyberweapons, making cancerous lesions appear in imaging studies where none exist. Recent research published in the European Journal of Radiology, however, suggests that radiologists still have the upper hand and can distinguish real lesions from AI-generated ones.

Researchers from Switzerland’s University Hospital Zürich used a special type of AI program called a generative adversarial network (GAN) to see if they could fool radiologists.  The GAN algorithm consists of two distinct networks: a generator network that can add a false lesion to different images and a discriminator network that is programmed to detect those malicious additions. The two networks act in competition, trying to learn from the other in order to improve their own performance. In theory, with enough time and a large sample size of images, the generator should be able to fool the discriminator. But what the researchers did not know was whether the generator could also fool trained radiologists.

To test the idea, the researchers trained a cycle-consistent GAN model (CycleGAN) on 860 mammographic images from more than 300 patients, and then tested the model using images of 302 images showing cancerous lesions as well as 590 normal scans. The researchers then had three radiologists read both original images, as well as those that had been modified by the GAN, in both a lower and higher resolution, and then had them rate whether a suspicious lesion was present as well as their opinion regarding whether that lesion was a manipulated image using 1-to-5 scales.

The study authors discovered that, at a low resolution, the radiologists had difficulty distinguishing between original and modified images. At the higher resolution of 512 x 408 pixels, however, the radiologists not only showed a significantly lower detection rate on the modified images, but they also could more easily identify that they had been modified. The researchers concluded that CycleGAN can be used to insert suspicious features into existing images but is not yet a threat as a cyberweapon because of technical limitations.

Using AI to better detect vertebral fractures

Small vertebral fractures, which can be common in diseases like osteoporosis, are often missed by traditional imaging techniques. New research, published in Radiology, suggests a deep learning algorithm can assist with detection and diagnosis, helping clinicians identify these small breaks before they become a more significant medical issue.

Researchers from the University of Manitoba developed an AI algorithm called a convolutional neural network (CNN) to help identify these small fractures. The group trained and tested the CNN on more than 12,000 dual x-ray absorptiometry images to assess whether a fracture was present. They learned that the CNN was able to correctly detect vertebral fractures under the receiving operating characteristic curve of 0.94, with a sensitivity of 87.4% and a specificity of 88.4%. When compared to expert readers, with at least 10 years of experience in reading such images, the CNN had an agreement rate of 0.76, leading the researchers to concluded that such algorithms could augment radiologist’s abilities to detect vertebral fractures, helping to increase accuracy of diagnosis and improve patient outcomes in the future.

Should men get mammograms?

A new study suggests selective mammography screening for men at high-risk for the disease can result in earlier detection-and more lives saved. The research was published in Radiology.

Mammography has proved itself to be a valuable tool in detecting breast cancer in women. As incidence rates of breast cancer in males is rising, researchers from New York University Langone Medical Center, New York University School of Medicine, and New York University’s Center for Advanced Imaging Innovation and Research wondered if it would be as effective in identifying breast cancer in males as well. To date, there are no formal screening guidelines or recommendations for men, even those who have a family history of the disease or specific genetic mutations that put them at higher risk.

In a retrospective study, the researchers reviewed male breast imaging studies, as well as the disease outcomes, in nearly 2000 men, between the ages of 18 and 96 years, with a personal or family history of breast cancer or associated genetic mutations.  That review revealed that mammography helped to detect 2,304 breast lesions in those patients.  Of those, only 149 were biopsied, but 27.5% were shown to be malignant. Mammography screening yielded a cancer detection rate of approximately 18 per 1000 examinations. This is in stark contrast to women, where the average detection rate stands at approximately 3-5 cases per 1000 examinations in average risk women. Furthermore, mammography screening in men led to earlier detection of cancer, before the disease had spread, improving the chance of survival.

In addition, the researchers found the following:

• A mammographic screening sensitivity of 100% (95% CI: 50%, 100%);

• A mammographic screening specificity of 95% (95% CI: 93.1%, 98%);

• And a mammographic screening positive predictive value of biopsy at 50% (95% CI: 22.2%, 77.8%).

Taken together, the researchers concluded that there may well be benefit in screening men who meet high-risk criteria for developing breast cancer-and that more research should be undertaken to develop specific recommendations for doing so.

MRI may help diagnose minor stroke

If you’ve taken a CPR course, you’ve likely heard about the F.A.S.T. mnemonic. The idea is simple: if you see a person with face drooping (F), arm weakness (A), and speech difficulties (S), it’s time (T) to call emergency services as they are likely experiencing a stroke. Yet, not all strokes follow such a particular pattern of symptoms. Some patients may just experience numbness or dizziness, likely because it is only a minor stroke or transient ischemic attack (TIA), which may lead physicians to pursue an alternate diagnosis-and perhaps miss the window of opportunity for early treatment or identify the patient as being at higher risk for a later stroke.

New research in JAMA Neurology, however, suggests that emergency physicians may be able to rely on MRI to help them better identify minor stroke and TIAs.  The researchers wanted to establish the frequency of acute infarct detected by diffusion-weighted MRI scans. Across multiple hospital sites, 1,028 patients, 522 females and 506 males, with minor symptoms including nonmotor or nonspeech minor focal neurologic events were given a thorough neurological assessment as well as an MRI within 8 days of symptom onset. The researchers discovered:

• MRI results demonstrated that 139 patients (13.5%) had experienced an acute stroke, much higher than expected in such a low-risk population;

• The MRI results led physicians to revise their final diagnoses in 30% of patients;

• A DWI-positive MRI was associated with a higher risk of recurrent stroke one year later;

• Multivariate modeling showed that older age, male sex, motor or speech symptoms, ongoing symptoms at the time of assessment, and abnormal results for the initial neurological examination were also associated with MRI evidence of stroke.

The researchers concluded that MRI scans can offer predictive value in determining not only whether a person with unclear symptoms has experienced a stroke, but also if they are at higher risk for having one later in time.  As such, it should be considered when clinical assessment alone cannot reliably confirm or deny an ischemic episode.

CT may help detect peripheral prostate cancer

Screening for prostate adenocarcinoma in healthy men remains a controversial subject. To date, clinicians rely on digital rectal exams or a prostate-specific antigen test to identify early stage cancer. But a new study, published in the American Journal of Roentgenology, suggests that there may be diagnostic value in using CT scanning to detect cancer in the peripheral zone of the prostate.

Researchers from St. Michael’s Hospital’s Department of Medical Imaging in Toronto wanted to identify the sensitivity of contrast-enhanced CT in detecting prostate cancer. The study authors undertook a retrospective analysis of 100 patients with biopsy-proven prostate cancer who underwent a staged contrast-enhanced CT of the abdomen and pelvis within 3 months of diagnosis. They compared those scans to 100 randomly selected age-matched male controls with no history of prostate cancer who had also undergone similar CT scanning. They then recruited two blind readers to independently assess the likelihood of prostate adenocarcinoma based solely on the CT scans.

The blind readers correctly identified 83 out of the 100 patients with prostate cancer, as well as 93 of the 100 matched control participants with a sensitivity of 0.83, a specificity of 0.92, a PPV of 0.91, and NPV of 0.84). Interrater agreement was at 0.76. 

Based on these results, the study authors concluded that contrast-enhanced CT may allow radiologists to see a focal area of increased enhancement in the periphery of the prostate that has diagnostic value. Clinicians could use such scans to order further work-up, leading to earlier detection of this deadly disease.