Osseous Tumor Reporting and Data System—Multireader Validation Study
Chhabra, Avneesh MD, MBA; Gupta, Anurag BSE; Thakur, Uma MD; Pezeshk, Parham MD; Dettori, Nathan MD; Callan, Alexandra MD; Xi, Yin PhD, Weatherall, Paul MD
Journal of Computer Assisted Tomography
Background: To develop and validate an Osseous Tumor Reporting and Data System (OT-RADS) with the hypothesis that the proposed guideline is reliable and assists in separating benign from malignant osseous tumors with a good area under the curve, and that could assist further patient management.
Design: Multireader cross-sectional validation study
Methods : An agreement was reached for OT-RADS categories based on previously described magnetic resonance imaging features and consensus of expert musculoskeletal radiologists. World Health Organization classification was used, and a wide spectrum of benign and malignant osseous tumors was evaluated. Magnetic resonance imaging categories were as follows: OT-RADS 0—incomplete imaging; OT-RADS I—negative; OT-RADS II—definitely benign; OT-RADS III—probably benign; OT-RADS IV—suspicious for malignancy or indeterminate; OT-RADS V—highly suggestive of malignancy; and OT-RADS VI—known biopsy-proven malignancy or recurrent malignancy in the tumor bed. Four blinded readers categorized each tumor according to OT-RADS classification. Intraclass correlation (ICC) and Conger κ were used. Diagnostic performance measures including area under the receiver operating curve were reported. Osseous Tumor Reporting and Data System was dichotomized as benign (I–III) and malignant (IV and V) for calculating sensitivity and specificity.
Main Results: Interreader agreement for OT-RADS and binary distinction of benign versus malignant were good to excellent, while agreement for individual tumor feature characteristics were poor to fair. The sensitivities, specificities, and area under the receiver operating curve of the readers ranged from 0.93–1.0, 0.71–0.86, and 0.92–0.97, respectively.
Conclusions : Osseous Tumor Reporting and Data System lexicon is reliable and helps stratify tumors into benign and malignant categories. It can be practically used by radiologists to guide patient management, improve multidisciplinary communications, and potentially impact outcomes.
Trainees take home points:
- Tumors can be classified using multiple criteria.
- Osseous Tumor Reporting and Data System (OT-RADS) is a system which has been validated and used for classification.
OT-RADS has good inter reader agreement in ascertaining the benign and aggressive nature of osseous tumors.
Significance of Acquisition Parameters for Adipose Tissue Segmentation on CT Images
Amelie S. Troschel, MD, Fabian M. Troschel, MD, Georg Fuchs, Cand Med, J. Peter Marquardt, Cand Med, Jeanne B. Ackman, MD, Kai Yang, PhD and Florian J. Fintelmann, MD
https://www.ajronline.org/doi/10.2214/AJR.20.23280
American Journal of Roentgenology
Background: CT-based body composition analysis quantifies skeletal muscle and adipose tissue. However, acquisition parameters and quality can vary between CT images obtained for clinical care, which may lead to unreliable measurements and systematic error.
Questions: What is the influence of IV contrast medium, tube current–exposure time product, tube potential, and slice thickness on cross-sectional area (CSA) and mean attenuation of subcutaneous (SAT), visceral (VAT), and inter-muscular adipose tissue (IMAT)?
Design: Retrospective study
Methods: 244 images from 105 patients were analysed. We applied semiautomated threshold-based segmentation to CTA, dual-energy CT, and CT images acquired as part of PET examinations. An axial image at the level of the third lumbar vertebral body was extracted from each examination to generate 139 image pairs. Images from each pair were obtained with the same scanner, from the same patient, and during the same examination. Each image pair varied in only one acquisition parameter, which allowed us to estimate the effect of the parameter using one-sample t or median tests and Bland-Altman plots.
Main Results: IV contrast medium application reduced CSA in each adipose tissue compartment, with percentage change ranging from −0.4% to −9.3%. Higher tube potential reduced SAT CSA and VAT CSA and increased IMAT CSA. Thinner slices increased CSA in the VAT and IMAT compartments. Lower tube current–exposure time product had a variable effect on CSA. IV contrast medium and higher tube potential increased mean attenuation, with percentage change ranging from 0.8% to 1.7% and from 6.2% to 20.8%, respectively. Conversely, thinner slice and lower tube current–exposure time product reduced mean attenuation, with percentage change ranging from −5.4% to −1.0% and from −8.7% to −1.8%, respectively.
Conclusion: Acquisition parameters significantly affect CSA and mean attenuation of adipose tissue. Details of acquisition parameters used for CT-based body composition analysis need to be scrutinized and reported to facilitate interpretation of research studies.
Senior Editor– Thank you for highlighting the measurement errors caused by different CT parameters. Much attention has been focused on single slice segmentation for muscle and fat while definitions of sarcopenia are not clear at this point. Ideally, large volume (may be whole pelvis or abdomen) segmentations would be better to encompass the relevant anatomy and clinical correlatons, like done in our recent work- Gu P, Chhabra A, Chittajallu P, Chang C, Mendez D, Gilman A, Fudman DI, Xi Y, Feagins LA. Visceral Adipose Tissue Volumetrics Inform Odds of Treatment Response and Risk of Subsequent Surgery in IBD Patients Starting Antitumor Necrosis Factor Therapy. Inflamm Bowel Dis. 2021 Jul 22:izab167. doi: 10.1093/ibd/izab167. Epub ahead of print. PMID: 34291800.
Trainees take home points:
- CT based analysis of body compositions is useful in evaluation and qualification of skeletal muscle and adipose tissue. These measurements can be useful in the assessment of sarcopenia
- CT measurements can be impacted by : IV contrast medium, Higher & lower tube potential, Thinner slices
Fibular Tip Periostitis: New Radiographic Sign Predictive of Chronic Peroneal Tendon Subluxation-Dislocation in Pes Planovalgus
Valentino D. Abballe, MD, Mohammad Samim, MD, Elisa Ramos Gavilá, MD, William R. Walter, MD, Erin F. Alaia, MD and Zehava Sadka Rosenberg, MD
https://www.ajronline.org/doi/abs/10.2214/AJR.20.23964
American Journal of Roentgenology
Background: Tearing of the superior peroneal retinaculum is a known cause of peroneal tendon subluxation-dislocation. However, with the exception of cortical avulsions at the fibular attachment, superior peroneal retinaculum injury and subsequent peroneal tendon subluxation-dislocation are typically radiographically occult. Chronic undiagnosed peroneal tendon subluxation-dislocation can be a persistent cause of lateral ankle pain, leading to further degeneration and the possibility of complete peroneal tendon tears.
Objective: Is there an association between radiographic fibular tip periostitis and MRI evidence of peroneal tendon subluxation-dislocation in patients with hindfoot valgus?
Design: Retrospective study
Inclusion criteria: Thirty-five patients with radiographic fibular tip periostitis and 35 age and sex-matched individuals without periostitis were selected from among 220 consecutively registered patients with hindfoot valgus who had undergone both ankle radiography and MRI.
Methods: Studies were retrospectively assessed by two musculoskeletal radiologists in consensus and by two additional blinded radiologists independently for the presence of peroneal tendon subluxation-dislocation, presence of subfibular impingement, and hindfoot valgus angle measurements. Interobserver agreement and accuracy, sensitivity, and specificity in the detection of fibular periostitis, peroneal tendon subluxation-dislocation, and subfibular impingement were calculated, and Fischer exact, Mann-Whitney, kappa coefficient, and intraclass correlation tests were performed.
Main Results: Patients with hindfoot valgus and fibular tip periostitis were significantly more likely to have MRI evidence of peroneal tendon subluxation-dislocation (PTS) and subfibular impingement (62.9% and 68.6%, p < .001) than those without periostitis (5.7% and 0%, p < .001). Periostitis predicted PTS with 91.7% sensitivity and 71.7% specificity.
Drawing across axial plane show direct fibular and calcaneal contact due to calcaneal valgus with secondary stripping of superior peroneal retinaculum (open arrows) at its fibular attachment, distal fibular tip periostitis (solid arrow), and dislocation of peroneal tendons (arrowhead, B). Ca = calcaneus, F = fibula, T = talus. 2. 61-year-old woman with hindfoot valgus, distal fibular tip periostitis, peroneus longus tendon subluxation, and subfibular impingement.
(A) AP radiograph of ankle show hindfoot valgus and periostitis along lateral cortical margin of fibular tip (arrowheads). (B) Axial fat-suppressed proton-density MR image shows lateral subluxation of peroneus longus tendon at level of retromalleolar groove (arrow) and fibular periostitis (arrowhead). Asterisk indicates poor definition of superior peroneal retinaculum at attachment to fibula.
Conclusion: The radiographic finding of fibular tip periostitis in patients with hindfoot valgus can be a predictor of peroneal tendon subluxation-dislocation and may also suggest advanced hindfoot valgus and subfibular impingement. These radiographic associations should be recognized by the radiologist, and MRI may be recommended as clinically indicated.
Senior Editor: It is very interesting study considering chronic fibular enthesopathy is extremely common and ‘fleck sign’ is a known finding indicative of peroneal retinacular injury. The findings are also counterintuitive as peroneal tendon injuries/subluxations and 5th metatarsal fractures are more common in cavus-adductus type foot and not pes planus-abductovalgus type foot. A larger consecutive sample might be helpful to further study or validate these results.
One of the limitations they talk about is having used the keyword “ankle MRI” to search the cases, which introduces a significant selection bias in the study. A prospective study would thus be more beneficial.
Trainees take home points:
- Superior peroneal retinaculum stabilizes the peroneal tendon translation and function.
- “Fleck sign” is radiographic indications of avulsion of the superior peroneal retinacular avulsion
- Chronic peroneal tendon subluxation due to retinacular injury can lead to ankle pain and potential tearing of the tendon.
- Fibular tip periostitis on radiographs can suggest the presence of peroneal tendon subluxation.
Diffusion-Weighted MRI to Assess Sacroiliitis: Improved Image Quality and Diagnostic Performance of Readout-Segmented Echo-Planar Imaging (EPI) Over Conventional Single-Shot EPI
Hua Zhang, MD, Hongjie Huang, MD, Yuyang Zhang, MD, Zhanhai Tu, MD, Zebin Xiao, MD, Junmin Chen, MD, PhD and Dairong Cao, MD
https://www.ajronline.org/doi/10.2214/AJR.20.23953
American Journal of RoentgenologyBackground: DWI using single-shot echo-planar imaging (ss-EPI) is prone to artifacts, signal-intensity dropout, and T2* blurring. Readout-segmented echo-planar imaging (rs-EPI) may improve image quality in DWI of the sacroiliac joints.
Objective: The purposes of this study were, first, to qualitatively and quantitatively compare image quality between ss-EPI and rs-EPI DWI of the sacroiliac joints; and, second, to evaluate whether ADC values derived from ss-EPI and rs-EPI can differentiate disease activity in patients with axial spondyloarthritis (axSpA).
Design: Retrospective study
Inclusion criteria: The study included 75 patients who underwent ss-EPI and rs-EPI DWI of the sacroiliac joints.
Methods: Patients were classified into axSpA (n = 50) and no-ax-SpA (n = 25) groups on the basis of Assessment of SpondyloArthritis International Society (ASAS) criteria. Patients in the axSpA group were assigned to one of four disease activity states using the Ankylosing Spondylitis Disease Activity Score–C-reactive protein (ASDAS-CRP). Two radiologists independently assessed qualitative (overall image quality and diagnostic confidence) and quantitative (ADC, signal-to-noise ratio [SNR], and contrast-to-noise ratio [CNR]) imaging parameters.
Main Results: Readout-segmented EPI provided significantly better overall image quality, diagnostic confidence, SNR, and CNR than ss-EPI. In patients with axSpA, the correlation coefficients (r) of ADC values and ASDAS-CRP values were 0.456 and 0.458 for ss-EPI and 0.537 and 0.558 for rs-EPI. ADCs showed progressive increases with increasing activity state for both sequences, although these increases were more substantial for rs-EPI than for ss-EPI. ADC values for ss-EPI were significantly different only between the inactive and very high disease activity groups. ADC values for rs-EPI were significantly different between the inactive and high, inactive and very high, as well as the moderate and very high disease activity groups (p < .0083, Bonferroni-corrected threshold).
26-year-old man with axial spondyloarthritis. Coronal oblique fat-suppressed T2-weighted image (A) shows bilateral bone marrow edema of sacroiliac joints.
DW image (b = 600 s/mm2) obtained with single-shot echo-planar imaging (ss-EPI) (B) shows more distortion artifacts on sacroiliac joints than DW image obtained with readout-segmented echo-planar imaging (rs-EPI) (C). Lesions with high signal intensity can be visualized more clearly on rs-EPI than on ss-EPI.
Conclusion: Readout-segmented EPI significantly improves the image quality of DWI in imaging the sacroiliac joints. In patients with axSpA, activity states are better differentiated by rs-EPI than by ss-EPI. Readout-segmented EPI is a more robust tool than ss-EPI for imaging of axSpA and should be included in routine clinical protocols for MRI of the sacroiliac joints.
Senior editor: Thank you for an interesting work. While the work validates the image technical improvements, I find it very interesting that advantages of DWI have also been shown for axSpA lesions. In editor’s experience, DWI is also very useful in qualitative demonstration of soft tissue enthesitis if intravenous contrast has not been administered due to its excellent sensitivity and background signal and fat signal suppression.
Trainees take home points:
- Readout-segmented ecoplanar imaging improves the image quality of DWI for sacroiliac joint imaging.
- Readout- segmented ecoplanar imaging is better than single-shot echo-planar imaging for axial spondyloarthritis imaging.
Minimally Invasive Ultrasound-Guided Carpal Tunnel Release Improves Long-Term Clinical Outcomes in Carpal Tunnel Syndrome
Sarah I. Kamel, MD, Bradley Freid, BS, Corbin Pomeranz, MD, Ethan J. Halpern, MD and Levon N. Nazarian, MD
https://www.ajronline.org/doi/10.2214/AJR.20.24383
American Journal of Roentgenology
Background: Ultrasound guidance allows performance of carpal tunnel release with smaller incisions and quicker recovery than traditional open or endoscopic surgery.
Question: What is the long-term effectiveness of ultrasound-guided carpal tunnel release in improving function and discomfort in patients with carpal tunnel syndrome?
Design: Retrospective study
Participants: 61 ultrasound-guided carpal tunnel release procedures performed on 46 patients (25 women, 21 men; mean age, 60.6 years; range, 21–80 years) (15 bilateral procedures) with clinically diagnosed carpal tunnel syndrome.
Methods: The procedures were performed with a single-use transection device and local anesthesia at an outpatient radiology office. Patients answered three questionnaires (Quick Disabilities of the Arm, Shoulder, and Hand [QDASH] and two parts of the Boston Carpal Tunnel Syndrome Questionnaire—the symptom severity [BCTSQ-SS] and functional status [BCTSQ-FS] scales) to assess the function of and discomfort in the affected wrist immediately before and 2 weeks and at least 1 year after the procedure. Higher scores indicated increasing disability. Patients also answered a global satisfaction question at follow-up. Preprocedure and postprocedure scores were compared by paired Wilcoxon signed rank tests.
Main Results: The 46 patients had median preprocedure scores of 45.4 for QDASH, 3.2 for BCTSQ-SS, and 2.5 for BCTSQ-FS. The median scores 2 weeks after the procedure were 22.5 for QDASH, 1.7 for BCTSQ-SS, and 1.9 for BCTSQ-FS. All scores decreased from preprocedure scores and surpassed reference standards for clinically important difference in scores. Follow-up questionnaires obtained for 90% of wrists a median of 1.7 years after the procedure showed further declines in median scores: 2.3 for QDASH, 1.2 for BCTSQ-SS, and 1.1 for BCTSQ-FS. At long-term follow-up evaluation, 96% of wrists had lower QDASH and 98% had lower BCTSQ (average of BCTSQ-SS and BCTSQ-FS) scores compared with the preprocedure scores. Among the patients who participated in the survey, 93% were satisfied or very satisfied with the long-term outcomes. No immediately postoperative complications occurred. Two patients needed surgical intervention 8 and 10 days after surgery, one for infection after injury and one for posttraumatic compartment syndrome.
Conclusion: Ultrasound-guided carpal tunnel release quickly improves hand function and reduces hand discomfort; improvement persists beyond 1 year.
Ultrasound-guided carpal tunnel release may be a safe, effective, and less invasive alternative to traditional surgery.
Senior editor: Great work application of a minimally invasive approach in an outpatient setting. Need more of such works to make invasive procedures cost-efficient and convenient for the patients.
Trainees take home points:
- Carpal tunnel release is a surgical treatment option for patients suffering from carpal tunnel symptoms.
- Ultrasound guided release adds a minimally invasive guided option to perform a carpal tunnel release.
- Ultrasound guided release option provides another tool for the interventional musculoskeletal radiologists in treating carpal tunnel patients.
Development and Validation of a Radiomics Model for Differentiating Bone Islands and Osteoblastic Bone Metastases at Abdominal CT
Ji Hyun Hong, Joon-Yong Jung , Aram Jo, Yoonho Nam, Seongyong Pak, So-Yeon Lee, Hyerim Park, Seung Eun Lee, Sanghee Kim
https://pubs.rsna.org/doi/10.1148/radiol.2021203783
Radiology
Background: A CT radiomics–based random forest model was proven useful for differentiating bone islands from osteoblastic metastases at abdominal CT and showed better diagnostic performance compared with an inexperienced radiologist. It is important to diagnose sclerotic bone lesions in order to determine treatment strategy.
Question: How effective is a CT radiomics–based machine learning model for differentiating bone islands and osteoblastic bone metastases?
Design: Retrospective study
Inclusion criteria: Patients who underwent contrast-enhanced abdominal CT and were diagnosed with a bone island or osteoblastic metastasis between 2015 to 2019 at either of two different institutions were included: institution 1 for the training set and institution 2 for the external test set.
Participants: A total of 177 patients (89 with a bone island and 88 with metastasis; mean age, 66 years ± 12 [standard deviation]; 111 men) were in the training set, and 64 (23 with a bone island and 41 with metastasis; mean age, 69 years ± 14; 59 men) were in the test set.
Methods: Radiomics features were extracted. The random forest (RF) model was built using 10 selected features, and subsequent 10-fold cross-validation was performed. In the test phase, the RF model was tested with an external test set. Three radiologists reviewed the CT images for the test set. The sensitivity, specificity, accuracy, and area under the receiver operating characteristic curve (AUC) were calculated for the models and each of the three radiologists. The AUCs of the radiomics model and radiologists were compared.
Main Results: Radiomics features (n = 1218) were extracted. The average AUC of the RF model from 10-fold cross-validation was 0.89 (sensitivity, 85%; specificity, 82%; and accuracy, 84%). In the test set, the AUC of the trained RF model was 0.96 (sensitivity, 80%; specificity, 96%; and accuracy, 86%). The AUCs for the three readers were 0.95, 0.96, and 0.88. The AUC of radiomics model was higher than that of only reader 3 (0.96 vs 0.88, respectively; P = .03).
Graph shows the receiver operating characteristic curve of the radiomics-based random forest model and those of each of the three readers for differentiating bone islands and osteoblastic metastasis in the external test set. The area under the receiver operating characteristic curve (AUC) of the radiomics model was higher than that of reader 3 (0.96 vs 0.88, respectively; P = .03) and comparable to those of readers 1 and 2 (P = .81 and P = .88, respectively).
Conclusion: A CT radiomics–based random forest model was proven useful for differentiating bone islands from osteoblastic metastases and showed better diagnostic performance compared with an inexperienced radiologist.
Senior editor: Interesting study. Good use case of AI. Generally, features of density similar to bony cortex, spiculated margins of lesion, small size <1cm, elongated shape adjacent to joint with orientation along the long axis of bone are excellent diagnostic criteria for a bone island versus a sclerotic metastasis. I wonder if inexperienced radiologists are taught such criteria, how many would overcall the benign lesions as metastasis.
Trainees take home points:
- Radiomics and AI tools are gradually improving and proving to be value tools in image interpretation.
- With improved training models and robust data sets , the AI tools will be able to help identify lesions with more accuracy.
- In this paper, CT Radiomics can help identify and differentiate bone metastatic lesions from bone islands
Five-Minute Five-Sequence Knee MRI Using Combined Simultaneous Multislice and Parallel Imaging Acceleration: Comparison with 10-Minute Parallel Imaging Knee MRI
Filippo Del Grande, Ali Rashidi, Rodrigo Luna, Marco Delcogliano, Steven E. Stern, Danoob Dalili, Jan Fritz
https://pubs.rsna.org/doi/abs/10.1148/radiol.2021203655?journalCode=radiology
Radiology
Background: Rapid knee MRI using combined simultaneous multislice (SMS) technique and parallel imaging (PI) acceleration can add value through reduced acquisition time but requires validation of clinical efficacy.
Purpose: To evaluate the performance of clinical fourfold SMS-PI–accelerated, 5-minute, five-sequence, multicontrast knee MRI protocols compared with standard twofold PI-accelerated, 10-minute knee MRI protocols.
Design: Adults with painful knee conditions were prospectively enrolled from April 2018 to October 2019.
Inclusion criteria: An agreement to participate, age of 18 years or older, the ability to position the knee in MRI coils, the ability to remain in the supine position for at least 20 minutes, and the presence of a knee condition associated with pain and dysfunction that qualified for an indication for the knee.
Participants: A total of 252 adults were evaluated (mean age ± standard deviation, 47 years ± 17; 134 men).
Methods: Participants underwent fourfold SMS-PI–accelerated, 5-minute, turbo spin-echo (TSE) knee MRI and standard-of-care twofold PI-accelerated, 10-minute, TSE knee MRI at either 1.5 T or 3.0 T. Three radiologists independently evaluated the knee MRI studies for meniscal, tendinous, ligamentous, and osseocartilaginous injuries. Statistical analyses included k-based intermethod agreements and diagnostic performance testing. P < .05 was considered indicative of a statistically significant difference.
Main Results:
- Combined simultaneous multislice (SMS) technique MRI and parallel imaging (PI) enables fourfold-accelerated 5-minute five-sequence turbo spin-echo knee MRI at 1.5 T and 3.0 T.
- Five-minute SMS-PI–accelerated and 10-minute PI-accelerated knee MRI had high reader agreements for detecting structural abnormalities at 1.5 T (κ >0.71 [95% CI: 0.56, 0.83]) and 3.0 T (κ >0.85 [95% CI: 0.69, 0.96]).
- Five-minute and 10-minute MRI protocols had similar diagnostic performances for internal derangement (areas under the receiver operating characteristic curve >0.78; P > .32).
Conclusion: Comparisons of 5-minute five-sequence simultaneous multislice– and parallel imaging (PI)–accelerated and 10-minute five-sequence PI-accelerated turbo spin-echo MRI of the knee suggest similar performances at 1.5 and 3.0 T.
Senior editor: Thanks for an excellent work in a large sample demonstrating that hardware and software improvements continue to push the boundaries of MR image acquisition and can render it comfortable for patients to receive their imaging in a quicker fashion. We are not far from a 3D Dixon type 4-minute volume sequence using compressed sense, which can produce any desired plane of image in different tissue contrasts and can display structures of interest in their orientations for proper assessment. Orthopedic community will need to embrace such images as well for its successful implementation.
Trainees take home points:
- Simultaneous multi-slice and parallel imaging can help accelerate the acquisition of MRI images and reduce patient time on an MRI magnet. Resulting in better patient compliance and improved thorough put through the MRI magnet.
- Such short time acquisitions using these newer techniques can provide good diagnostic quality images.
- It is important to expose the Orthopedic community to these images and newer sequences, as highlighted by the Senior Editor in the above statement.
Assessment of an AI Aid in Detection of Adult Appendicular Skeletal Fractures by Emergency Physicians and Radiologists: A Multicenter Cross-sectional Diagnostic Study
Loïc Duron , Alexis Ducarouge, André Gillibert, Julia Lainé, Christian Allouche, Nicolas Cherel, Zekun Zhang, Nicolas Nitche, Elise Lacave, Aloïs Pourchot, Adrien Felter, Louis Lassalle, Nor-Eddine Regnard, Antoine Feydy
https://pubs.rsna.org/doi/abs/10.1148/radiol.2021203886?journalCode=radiology
Radiology
Background: The interpretation of radiographs suffers from an ever-increasing workload in emergency and radiology departments, while missed fractures represent up to 80% of diagnostic errors in the emergency department.
Question: How effective is artificial intelligence (AI) in aiding radiologists and emergency physicians in the detection and localization of appendicular skeletal fractures?
Design: Cross-sectional Diagnostic study
Methods: The AI system was previously trained on 60,170 radiographs obtained in patients with trauma. The radiographs were randomly split into 70% training, 10% validation, and 20% test sets. Between 2016 and 2018, 600 adult patients in whom multi-view radiographs had been obtained after a recent trauma, with or without one or more fractures of shoulder, arm, hand, pelvis, leg, and foot, were retrospectively included from 17 French medical centers. Radiographs with quality precluding human interpretation or containing only obvious fractures were excluded. Six radiologists and six emergency physicians were asked to detect and localize fractures with (n = 300) and fractures without (n = 300) the aid of software highlighting boxes around AI-detected fractures. Aided and unaided sensitivity, specificity, and reading times were compared by means of paired Student t tests after averaging of performances of each reader.
Main Results:
- The artificial intelligence (AI) aid, which highlighted potential fractures on full-resolution radiographs, improved the sensitivity (8.7% increase, P = 0.006) and specificity (4.1% increase, P = 0.03) of emergency doctors and radiologists in the diagnosis of appendicular fractures.
- The stand-alone area under the receiver operating characteristic curve, requiring that the AI system detect the precise locations of all fractures on an examination, was 0.94 with a newer release of the AI system.
Conclusion: The artificial intelligence aid provided a gain of sensitivity (8.7% increase) and specificity (4.1% increase) without loss of reading speed.
The artificial intelligence aid improved the sensitivity and specificity of radiologists and emergency physicians in the localization of appendicular fractures on radiographs, with no additional reading time.
Senior editor: Excellent work demonstrating the utility of AI in repetitive and focused tasks, which can make the lives of treating physicians easier and a model/process that can reduce waiting time for doctors looking for a radiologist’s report.
Trainees take-home points:
- AI can help in aiding diagnosis
- Training the AI engine with various datasets and validation of the training will impact outcomes.
- Significant strides have been made in improving AI diagnostic aid and highlighting of injuries.
- AI will be able to help physicians in centers with limited timely radiologist input and improve the level of imaging-driven care.
- This paper demonstrates the role of AI in fracture detection.
