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

Best Of EAST 2023 #9: CT Of The Lumbar Spine

It is becoming clearer and clearer that patients with suspicion for fractures of the thoracic (T) and lumbar (L) spine should be imaged only with CT scan. Conventional imaging just doesn’t have enough sensitivity, even in younger patients with healthy bones. But when we obtain CT of the T&L spines there is a choice: just look at the axial / helical slices, or have the computer reconstruct additional images in the sagittal and coronal planes. The belief is that this multiplanar imaging will assist in finding subtle fractures that might not be seen on axial views.

The group at Rutgers in New Jersey tried to determine if adding the reconstructions amounted to overkill. They performed a retrospective review of patients at their Level I center over a six-year period. They focused on studies performed in patients who had T and/or L fractures who also had both CT of the chest, abdomen, and pelvis (CAP) and thoracic and lumbar reconstructions. Additional data were obtained from a review of the medical record and trauma registry.

Here are the factoids:

  • A total of 494 patients had both CAP and reconstructions
  • There were 1254 fractures seen on CAP, and an additional 129 fractures seen with recons (total of 1394)
  • The majority of additional injuries not detected on CT CAP were transverse process fractures
  • The number of other fracture patterns not seen on CT CAP were statistically “not significant”
  • However, these numerically “not significant” fractures included 51 vertebral body fractures, 6 burst fractures, 3 facet fractures, and 2 pedicle fractures
  • No unstable fractures were missed on CT CAP
  • More MRIs were performed in the patients who had recons, there were more spine consultations, and 11% underwent operative fixation vs. 2% for CTA only (!!)

The authors concluded that CT CAP alone was sufficient to identify clinically significant thoracic and lumbar fractures. They also stated that clinically insignificant injuries identified with reconstructions were more likely to undergo MRI and use excess resources. They urged us to be selective with the use of T&L reformats.

Bottom line: Wow! I have a lot of questions about this abstract! And I really disagree with the findings.

You studied fewer than 500 patients with T or L spine fractures over a six year period. This is only about 80 per year, which seems very low. This suggests that many, many patients were being scanned without recons to start with. How did patients get selected out to get recons? Were there specific criteria? I worry that this could add some bias to your study.

The number of fractures seen only on the recon views besides transverse process fractures were deemed “statistically insignificant.” However, looking at the list of them (see bullet point 5 above) they don’t look clinically insignificant. It’s no wonder that recons resulted in more consults, MRI scans, and spine operations!

I worry that your conclusion is telling us to stop looking for fractures so we won’t use so many additional resources. But their use may be in the best interest of the patients!

Here are my questions and comments for the presenter/authors:

  • Why did you decide to do this study? I didn’t realize that not doing the recons was a thing in major blunt trauma. Was there some concern that resources were being wasted? Was there an additional cost for the reconstructions?
  • How many patients only received CT CAP? The greater the number of these, the higher the probability that some non-random selection process is going on that might bias your findings.
  • How did you get separate reports for the non-reconstructed images? Did you have new reads by separate radiologists? Typically, the report contains the impression for the entire study. It would be unusual for the radiologist to comment on the non-recon images, then add additional findings from just the reconstructions.
  • Doesn’t the increased numbers of spine consults, MRIs, and operative procedures in the patients with reconstructions imply that these otherwise occult fractures needed clinically important additional attention? 

I worry that readers of this abstract might take away the wrong message. Unless there is some additional compelling data presented, this study is certainly not enough to make me change my practice!

Reference: UTILITY OF CT THORACOLUMBAR SPINAL RECONSTRUCTION IMAGING IN BLUNT TRAUMA. EAST 2023 Podium paper #20.

Print Friendly, PDF & Email

Best Of EAST 2023 #8: Use Of AI To Detect Rib Fractures On CT

Artificial intelligence systems (AI) are increasingly finding their way into medical practice. They have been used to assist pathologists in screening microscope specimens for years. Although still amazingly complicated, one of the most obvious applications for trauma is in reading x-rays. Counting rib fractures may be helpful for care planning, and characterizing fracture patterns may assist our orthopedic colleagues in evaluating and planning rib plating procedures.

The trauma group at Stanford developed a computer vision system to assist in identifying fractures and their percent displacement.  They used a variation on a neural network deep learning system and trained it on a publicly available CT scan dataset.  They used an index of radiographic similarity (DICE score) to test how well their model matched up against the reading of an actual radiologist.

Here are the factoids:

  • The AI network was trained on a dataset of 5,000 images in 660 chest CT scans that had been annotated by radiologists
  • The model achieved a DICE score of 0.88 after training
  • With a little jiggering of the model (reweighting), the receiver operating characteristic curve improved to 0.99, which is nearly perfect

The left side shows a CT scan rotated 90 degrees; the right side shows the processed data after a fracture was detected.

Bottom line: This paper describes what lies ahead for healthcare in general. The increasing sophistication and accuracy of AI applications will assist trauma professionals in doing their jobs better. But rest easy, they will not take our jobs anytime soon. What we do (for the most part) takes very complex processing and decision making. It will be quite some time before these systems can do anything more that augment what we do.

Expect to see these AI products integrated with PACS viewing systems at some point in the not so distant future. The radiologist will interpret images in conjunction with the AI, which will highlight suspicious areas on the images as an assist. The radiologist can then make sure they have reported on all regions that both they and the AI have flagged.

Here are my questions and comments for the presenter/authors:

  • How can you be sure that your model isn’t only good for analyzing your training and test datasets? If neural networks are overtrained, they get very good at the original datasets but are not so good analyzing novel datasets. Have you tried the on your own data yet?
  • Explain what “class reweighting” is and how it improved your model. I presume you used this technique to compensate for the potential issue mentioned above. But be sure to explain this in simple terms to the audience.
  • Don’t lose the audience with the net details. You will need to give a basic description of how deep learning nets are developed and how they work, but not get too fancy.

This is an interesting glimpse into what is coming to a theater near you, so to speak. Expect to see applications appearing in the next few years.

Reference: AUTOMATED RIB FRACTURE DETECTION AND CHARACTERIZATION ON COMPUTED TOMOGRAPHY SCANS USING COMPUTER VISION. EAST 2023 Podium paper #16.

Print Friendly, PDF & Email

Best Of EAST 2023 #7: The CT Autopsy

Back in the day, autopsy after trauma death was fairly commonplace. Nowadays, it is typically reserved for fatalities that involve a potential crime. And it can be challenging to get the medical examiner to release copies for trauma performance improvement.

One potential remedy for this began to surface in the literature about twenty years ago: the virtual (or CT) autopsy. This entails sending the postmortem patient to the scanner for head, cervical, chest, and pelvic scans. Although it seems like an exciting idea, there are several logistical issues that I will discuss later.

The trauma group at Indiana University performed a retrospective study to determine the common injury patterns in patients who died at or up to one hour after ED arrival. Their goal was to identify injury patterns that might improve the focus and quality of resuscitative efforts in living patients. They reviewed their experience with doing postmortem CT over a nine-year period. The primary goal was to identify sources of hemorrhage, TBI, and cervical spine injury. They also wanted to identify significant pneumothorax and misplaced airway devices.

Here are the factoids:

  • There were 80 decedents in the study, and they were severely injured, with an average ISS of 42
  • About three quarters arrested prior to arrival, and the remainder arrived with a pulse
  • The most common major injuries were severe TBI (41%), long bone fractures (25%), hemoperitoneum (23%), and cervical spine injury (19%)
  • A moderate pneumothorax was present in 19% of cases
  • Misplaced airway was identified in 5%
  • There was no difference in injury or device mishap patterns between pre-hospital and in-hospital arrest patients (although the number of patients was probably too small to detect one)

The authors concluded that the injury patterns between those who died prior to arrival vs. after were the same. They also noted that patients in arrest should automatically have their chest decompressed and the airway position checked.

Bottom line: This is an intriguing study of a concept I’ve been thinking about for years. The quality improvement benefits could be amazing! Imagine getting immediate feedback on the cause of death and how it might influence future resuscitations. The authors pointed out the power of this with their discovery of missed pneumothorax and malpositioned airways.

But, as mentioned above, there are a host of logistical problems to work out first. Here is a partial list:

  • Who accompanies the patient to scan? A nurse? The team?
  • Covered or uncovered? It might be creepy for people in the hallways to see a covered person being wheeled around. That’s why hospitals always have those white, wheeled boxes. But it’s equally creepy to see a person who is not moving or breathing being transported.
  • Be prepared for your radiologists to gripe about doing free reads
  • Where does the report go? It shouldn’t go to the medical record. Or should it?
  • What about liability issues? If the team misses something big and the report goes to the chart, it’s fair game for a lawsuit.
  • And many more!

Here are my questions and comments for the presenter/authors:

  • How did you come to do this study? It appears that your group has been performing CT autopsies for almost a decade. Was there a protocol? Was it done on every eligible patient? If not, could this have skewed your results?
  • Do you have the statistical power to detect any differences between the various groups? A few of your results did approach significance. Perhaps more subjects would have helped.
  • Tell us how you have addressed the logistical problems above.

This is great work; perhaps it will stimulate a move toward embracing this concept!

Reference: CHARACTERIZATION OF FATAL BLUNT INJURIES USING POST-MORTEM COMPUTED TOMOGRAPHY. EAST 2023 Podium paper #14.

 

Print Friendly, PDF & Email

Best Of EAST 2023 #5: Imaging The Elderly

Several papers have been published over the years regarding underdiagnosis when applying the usual imaging guidelines to elderly trauma patients. Unfortunately, our elders are more fragile than the younger patients those guidelines were based on, leading to injury from lesser mechanisms. They also do not experience pain the same way and may sustain serious injuries that produce no discomfort on physical exam. Yet many trauma professionals continue to apply standard imaging guidelines that may not apply to older patients.

EAST sponsored a multicenter trial on the use of CT scans to minimize missed injuries. Eighteen Level I and Level II trauma centers prospectively enrolled elderly (age 65+) trauma patients in the study over one year. Besides the usual demographic information, data on physical exams, imaging studies, and injuries identified were also collected. The study sought to determine the incidence of delayed injury diagnosis, defined as any identified injury that was not initially imaged with a CT scan.

Here are the factoids:

  • Over 5,000 patients were enrolled, with a median age of 79
  • Falls were common, with 65% of patients presenting after one
  • Nearly 80% of patients actually sustained an injury (!)
  • Head and cervical spine were imaged in about 90% of patients, making them the most common initial studies
  • The most commonly missed injuries involved BCVI (blunt carotid and vertebral injury) or thoracic/lumbar spine fractures
  • 38% of BCVI injuries and 60% of T/L spine fractures were not identified during initial imaging
  • Patients who were transferred in, did not speak English, or suffered from dementia were significantly more likely to experience delayed diagnosis

The authors concluded that about one in ten elderly blunt trauma patients sustained injuries in body regions not imaged initially. They recommended the use of imaging guidelines to minimize this risk.

Bottom line: Finally! It has taken this long to perform a study that promotes standardizing how we perform initial patient imaging after blunt trauma. Granted, this study only applies to older patients, but the concept can also be used for younger ones. The elderly version must mandate certain studies, such as head and the entire spine. Physical exams can  still be incorporated in the guidelines for younger patients but not the elderly.

The overall incidence of BCVI was low, only 0.7%. But its presence was missed in 38% of patients, setting them up for a potential  stroke. Some way to incorporate CT angiography of the neck will need to be developed. The risk / benefit ratio of the contrast load vs. stroke risk will also have to be determined.

Here are my questions and comments for the presenter/authors:

  • Did you capture all of the geriatric patients presenting to the study hospitals? By my calculation, 5468 patients divided by 18 trauma centers divided by 14 months of study equals 22 patients enrolled per center per month. Hmm, my center sees more than that number of elderly injured patients in the ED per day! Why are there so few patients in your study? Were there some selection criteria not mentioned in the abstract?
  • Why should we believe these study numbers if you only included a subset of the total patients that were imaged?

My own reading of the literature leads me to believe that your conclusions are correct. I believe that all centers should develop or revise their elderly imaging guidelines to include certain mandatory scans regardless of how benign the physical exam appears. Our elders don’t manifest symptoms as reliably as the young. But the audience needs a little more information to help them understand some of the study numbers.

Reference: SCANNING THE AGED TO MINIMIZE MISSED INJURY, AN EAST MULTICENTER TRIAL. EAST 2023 podium abstract #12.

Print Friendly, PDF & Email