Category Archives: Imaging

New Technology: Using AI To Interpret Pelvic X-rays

Look out, radiologists! The computers are coming for you!

Radiologists use their extensive understanding of human anatomy and combine it with subtle findings they see on x-ray shadow pictures. In doing this, they can identify a wide variety of diseases, anomalies, and injuries. But as we have seen with vision systems and game playing (think chess), computers are getting pretty good at doing this as well.

Is it only a matter of time until computer artificial intelligence (AI) starts reading x-rays?  Look at how good they already are at interpreting EKGs. The trauma group at Stanford paired up with the Chang Gung Memorial Hospital in Taiwan to test the use of AI for interpreting images to identify a specific set of common pelvic fractures.

The Stanford group used a deep learning neural network (XCeption) to analyze source x-rays (standard A-P pelvis images) from Chang Gung. These x-rays were divided into training and testing cohorts. The authors also applied different degrees of blurring, brightness, rotation, and contrast adjustment to the training set in order to help the AI overcome these issues when interpreting novel images.

The AI interpreted the test images with a very high degree of sensitivity, specificity, accuracy, and predictive values, with all of them over 0.90. The algorithms generated a “heat map” that showed the areas that were suspicious for fracture. Here are some examples with the original x-ray on the left and the heat map on the right:

The top row shows a femoral neck fracture, the middle row an intertrochanteric fracture, and the bottom row another femoral neck fracture with a contralateral implant. All were handily identified by the AI.

AI applications are usually only as good as their training sets. In general, the bigger the better so they can gain a broader experience for more accurate interpretation. So it is possible that uncommon, subtle fractures could be missed. But remember, artificial intelligence is meant to supplement the radiologist, not replace him or her. You can all breathe more easily now.

This technology has the potential for broader use in radiographic interpretation. In my mind, the best way to use it is to first let the radiologist read the images as they usually do. Once they have done this, then turn on the heat map so they can see any additional anomalies the AI has found. They can then use this information to supplement the initial interpretation.

Expect to see more work like this in the future. I predict that, ultimately, the picture archiving and communications systems (PACS) software providers will build this into their product. As the digital images are moving from the imaging hardware to the digital storage media, the AI can intercept it and begin the augmented interpretation process. The radiologist will then be able to turn on the heat map as soon as the images arrive on their workstation.

Stay tuned! I’m sure there is more like this to come!

Reference: Practical computer vision application to detect hip fractures on pelvic X-rays: a bi-institutional study.  Trauma Surgery and Acute Care Open 6(1), http://dx.doi.org/10.1136/tsaco-2021-000705.

Print Friendly, PDF & Email

Arms Up or Arms Down In Torso CT Scans?

CT scan is a valuable tool for initial screening and diagnosis of trauma patients. However, more attention is being paid to radiation exposure and dosing. Besides selecting patients carefully and striving for ALARA radiation dosing (as low as reasonably achievable) by adjusting technique, what else can be done? Obviously, shielding parts of the body that do not need imaging is simple and effective. But what about simply changing body position?

One simple item to consider is arm positioning in torso scanning. There are no consistent recommendations for use in trauma scanning. Patients with arm and shoulder injuries generally keep the affected upper extremity at their side. Radiologists prefer to have the arms up if possible to reduce scatter and provide clearer imaging.

Radiation physics research has examined arm positioning and its effect on radiation dose. A retrospective review of 690 patients used dose information computed by the CT software and displayed on the console. Radiation exposure was estimated using this data and was stratified by arm positioning. Even though there are some issues with study design, the results were impressive.

The dose results were as follows:

  • Both arms up: 19.2 mSv (p<0.0000001)
  • Left arm up: 22.5 mSv
  • Right arm up: 23.5 mSv
  • Arms down: 24.7 mSv

Bottom line: Do everything you can to reduce radiation exposure:

  1. Be selective with your imaging. Do you really need it?
  2. Work with your radiologists and physicists to use techniques that reduce dose yet retain image quality
  3. Shield everything that’s not being imaged.
  4. Think hard about getting CT scans in children
  5. Raise both arms up during torso scanning unless injuries preclude it.

Reference: Influence of arm positioning on radiation dose for whole body computed tomography in trauma patients. J Trauma 70(4):900-905, 2011.

Print Friendly, PDF & Email

Using Chest CT To Detect Occult Injuries

There are major belief systems when it comes to the use of trauma CT: selective scan vs pan scan. The selective scanners believe that too much radiation can be bad, and that the risk of excess exposure outweighs the value of scanning everything. The pan scanners believe that valuable information might be missed unless they routinely image everything.

Who is right? There’s probably value in each side of the argument. But do we have data? Good data? Two emergency medicine groups from UC-Irvine and UC-San Diego tried to answer this question via a prospective study involving 10 Level I trauma center EDs in California.  They tagged onto data collection underway for the NEXUS chest and chest CT studies from 2009-2012.

Patients with fresh (< 24 hours) blunt trauma who underwent chest imaging in the ED were included. Patients needed to have both CT scan and chest x-ray within 24 hours, at the discretion of the emergency physician. Weirdly, they skewed their sample by enrolling patients from 7am to 11pm daily due to availability of research personnel.

The researchers were looking for minor and major interventions necessitated by data discovered on the CT scan. Occult injuries were defined as clinically important if an intervention occurred because of it. Major interventions included surgery, mechanical ventilation for pulmonary contusion, or chest tube for hemo- and pneumothoraces.

Here are the factoids:

  • Nearly six thousand patients were enrolled, and 2,048 had at least one injury identified on either study
  • A total of 1,454 of these injuries (71% of injuries, but only 25% of patients) were occult, only being seen on the CT scan
  • Chest x-ray found all injuries in only 29% of patients (not surprising)
  • When pulmonary contusion was seen by CT only, 6% were placed on ventilators; when hemo- or pneumothorax were seen, 41% and 29% respectively had chest tubes inserted (wow!)
  • The authors tallied 241 major interventions for occult injury in 202 patients, 154 chest tubes for hemo/pneumothorax and/or mechanical ventilation, 9 operations for diaphragm or aortic injury, and the remainder appear to be for other chest wall fractures

The authors concluded that occult injuries were found in 71% of their patients, with the majority of those “requiring” chest tubes. They recognized some of the shortcomings in their study and stopped short of recommending a pan-scan type approach to major chest trauma.

Bottom line: This argument always boils down to diagnostic yield vs money vs radiation. Radiologists like to find as many things as they can, so CT is great. For me, it always comes back to that old saying: “if a tree falls in the woods when no one is around, does it make a sound?” 

The corollary is “if a diagnosis is found on CT that is not clinically relevant, do we care?” But wait, you say, they did have to intervene. Or did they?

Have you ever scanned a chest and seen something that makes you intubate the patient immediately and put them on a ventilator? Probably not. It’s a clinical judgement. The scan may make you a bit more wary, but you will still wait for some clinical signs that the patient needs that extra help. 

And what about chest tube insertion? I’m sure most of you have seen a modest pneumothorax on chest x-ray (1 cm away from the chest wall, extending to the 6th intercostal space, say). Ho hum. And then you get a CT scan and your eyes widen. It always looks much larger on the scan. It always does. Yet the patient is still lying there comfortably, with normal oxygen saturations. Do you really need to put a tube in? For decades, we used only the x-ray, and patients did fine.

So I don’t buy that the CT result made them do the interventions. It was the clinician’s choice based on how they interpreted the scan, not the patient’s clinical condition. Without specific guidelines that determine when interventions are indicated, it just boils down to “I do an intervention when I think the patient needs it.” And every clinician will have their own criteria and thresholds. It’s tough to learn from things done this way.

So I stick by my guns. We know that chest x-ray is flawed. But it does provide good clinical data even without a bunch of diagnostic minutiae. A good practice guideline that helps select the patients most likely to benefit from a CT scan is paramount.

As you can probably tell, I’m a selective scan kind of guy and still have not run across a study that is clean and compelling enough to make me change. And I think I’ll be waiting for a while for one of those to pop up!

Reference: Prevalence and clinical import of thoracic injury identified by chest computed tomography but not chest radiography in blunt trauma: multicenter prospective cohort study. Ann Emerg Med 66(6):589-600, 2015.

Print Friendly, PDF & Email

Best of EAST #3: Spine MRI Usage After EAST Guidelines

In 2015, EAST published their practice guidelines for spine clearance in the obtunded blunt trauma patient. Click here to view them. They stated that a high-quality CT scan can be used to remove (clear) the cervical collar in these patients. This avoids the use of the expensive and personnel-intensive MRI clearance.

The group at UCSF used the NTDB to review the use of MRI in such patients over an 11 year period. They focused on comatose patients (GCS < 8) with an AIS head > 3 and intubation for more than 72 hours. They used logistic regression to equalize confounders while examining the use of MRI over time, before and after the guidelines were published.

Here are the factoids:

  • More than 75,000 patients from 530 trauma centers were included
  • Patients who were older, Hispanic, uninsured, or involved in a car crash were less likely to undergo spinal MRI
  • Level I centers were more likely to use MRI for clearance than Level II centers
  • Patients evaluated after release of the practice guidelines were 1.7x more likely to undergo MRI for spine clearance (!!)

The authors concluded that spinal MRI use has been increasing since 2007 despite publication of the EAST guideline.

My comments: To me, this indicates one of the following:

  1. Nobody reads the EAST guidelines, or
  2. Trauma programs believe that they alone are able to figure out what is right, and everyone else is wrong

I suspect that it is #2. For some reason, trauma programs insist on doing it their own way despite what decent evidence shows. I think that this represents a defense mechanism to minimize the cognitive dissonance that comes with defying what is published in the literature.

I always encourage programs to borrow/steal what is already out there when crafting their own practice guidelines. Someone else has already done the work, why not take advantage of it? Typically, it’s just an excuse to continue doing things the way they’ve always been done.

This incessant reinventing the wheel becomes tiresome. And for once, I don’t have many questions or suggestions for the authors. Their evidence is pretty well laid out. 

My questions for the author / presenter are:

  1. Do you use MRI for spine clearance in your obtunded blunt trauma patients? And if so, WHY?
  2. Why do you think there are demographic and trauma center level disparities? Is it the teaching environment? Something else?

To everyone else, I say “get over yourself and read the literature!”

Reference: Assessing the e3ffect of the EAST guideline on utilization of spinal MRI in the obtunded adult blunt trauma patient over time. EAST 2021, Paper 7.

Print Friendly, PDF & Email

How To Detect Bucket-handle Intestinal Injuries With CT Scan

A bucket-handle injury is a relatively uncommon complication of blunt trauma to the abdomen. It only occurs in a few percent of patients, but is much more likely if they have a seat belt sign.  The basic pathology is that the bowel mesentery (small bowel of sigmoid colon) gets pulled away from the intestinal wall.

This injury is problematic because it may take a few days for the bowel itself to die and perforate. Patients with no other injuries could potentially be discharged from the hospital before they become overtly symptomatic, leading to delayed treatment.

Here’s an image from my personal collection with not one, but four bucket-handle injuries.

Typical patients with suspected blunt intestinal injury are observed with good serial exams and a daily WBC count. If this begins to rise after 24 hours, there is a reasonable chance that this injury is present.

CT scan has not really been that reliable in past studies. There may be some “dirty mesentery”, which is contused and has a hematoma within it. But without a more convincing exam, it is difficult to convince yourself to operate immediately on these patients.

A paper was published by a group of radiologists at Duke University. It appears to be a case report disguised as a descriptive paper. It looks like they picked a few known bucket-handle injuries from their institution and back-correlated them with CT findings.

The authors called out the usual culprits:

  • Fluid between loops of bowel
  • Active bleeding in the mesentery
  • Bowel wall perfusion defects

But they also noted that traumatic abdominal wall hernias were highly with the injury as well. These are rare, but should bring intestinal injury to mind when seen.

With newer scanners, radiologists are better able to detect subtle areas of hypoperfusion as well. This is a fairly good indicator of injury, especially when adjacent bowel appears normally perfused. Here are two examples. The black arrows denote active extravasation, and the white ones an area of hypoperfusion.

The authors add bowel wall hypoperfusion as another finding that may point to a bucket-handle type injury

Bottom line: Hold the phone! Don’t change your practice yet. This paper is not able to demonstrate how good this radiographic sign is. Looking at other radiology literature, the specificity is about 90%. But remember, that means that if they don’t have the CT finding, that’s true only 90% of the time.

Unfortunately the sensitivity is only 10%. So if you see it on the scan, they’ve got a 1 in 10 chance of actually having the injury. That’s not good enough for me to run to the operating room.

Here’s what I recommend: if your patient has an unconcerning exam and any of the usual culprits (pelvic fluid, inter-loop fluid, dirty mesentery, thickened bowel loops, abdominal wall hernia), perform serial exams and get a WBC the next morning. If the exam worsens, operate. If the WBC rises, consider laparoscopy to see if you need to make a bigger incision. And if you see this new kid on the block, the hypoperfused bowel, consider laparoscopy right away. 

I’m sure the radiologists and the technology will keep getting better. But for now, blunt intestinal injury still requires patience, perceptiveness, and a little luck.

References:

  • CT findings of traumatic bucket-handle mesenteric injuries. Am J Radiol 209:[email protected], 2017.
  • Multidetector CT of blunt abdominal trauma. Radiology 265(3):678–693, 2012.
Print Friendly, PDF & Email