Artificial intelligence for the classification of fractures around the knee in adults according to the 2018 AO/OTA classification system.
Ontology highlight
ABSTRACT: Background
Fractures around the knee joint are inherently complex in terms of treatment; complication rates are high, and they are difficult to diagnose on a plain radiograph. An automated way of classifying radiographic images could improve diagnostic accuracy and would enable production of uniformly classified records of fractures to be used in researching treatment strategies for different fracture types. Recently deep learning, a form of artificial intelligence (AI), has shown promising results for interpreting radiographs. In this study, we aim to evaluate how well an AI can classify knee fractures according to the detailed 2018 AO-OTA fracture classification system.Methods
We selected 6003 radiograph exams taken at Danderyd University Hospital between the years 2002-2016, and manually categorized them according to the AO/OTA classification system and by custom classifiers. We then trained a ResNet-based neural network on this data. We evaluated the performance against a test set of 600 exams. Two senior orthopedic surgeons had reviewed these exams independently where we settled exams with disagreement through a consensus session.Results
We captured a total of 49 nested fracture classes. Weighted mean AUC was 0.87 for proximal tibia fractures, 0.89 for patella fractures and 0.89 for distal femur fractures. Almost ¾ of AUC estimates were above 0.8, out of which more than half reached an AUC of 0.9 or above indicating excellent performance.Conclusion
Our study shows that neural networks can be used not only for fracture identification but also for more detailed classification of fractures around the knee joint.
SUBMITTER: Lind A
PROVIDER: S-EPMC8016258 | biostudies-literature |
REPOSITORIES: biostudies-literature
Similar Datasets
Project description:BACKGROUND:The classification systems for proximal humeral fractures routinely used in clinical practice include the Neer and Arbeitsgemeinschaft für Osteosynthesefragen/Orthopaedic Trauma Association (AO/OTA) 2007 systems. Currently used systems have low inter- and intraobserver reliability. In 2018, AO/OTA introduced a new classification system with the aim of simplifying the coding process, in which the Neer four-part classification was integrated into the fracture description. The aim of the present work is to assess the inter- and intraobserver agreement of the new AO/OTA 2018 compared with the Neer and AO/OTA 2007 classifications. MATERIALS AND METHODS:A total of 116 radiographs of consecutive patients with proximal humeral fracture were selected and classified by three observers with different levels of experience. All three observers independently reviewed and classified the images according to the Neer, AO/OTA 2007, and new AO/OTA 2018 systems. To determine the intraobserver agreement, the observers reviewed the same set of radiographs after an interval of 8 weeks. The inter- and intraobserver agreement were determined through Cohen's kappa coefficient analysis. RESULTS:The new AO/OTA 2018 classification showed substantial mean inter- (k = 0.67) and intraobserver (k = 0.75) agreement. These results are similar to the reliability observed for the Neer classification (interobserver, k = 0.67; intraobserver, k = 0.85) but better than those found for the AO/OTA 2007 system, which showed only moderate inter- (k = 0.57) and intraobserver (k = 0.58) agreement. The two more experienced observers showed better overall agreement, but no statistically significant difference was found. No differences were found between surgical experience and agreement regarding specific fracture types or groups. CONCLUSIONS:The results showed that the Neer system still represents the more reliable and reproducible classification. However, the new AO/OTA 2018 classification improved the agreement among observers compared with the AO/OTA 2007 system, while still maintaining substantial descriptive power and simplifying the coding process. The universal modifiers and qualifications, despite their possible complexity, allowed a more comprehensive fracture definition without negatively affecting the reliability or reproducibility of the classification system. LEVEL OF EVIDENCE:Level III, diagnostic studies.
| S-EPMC7067934 | biostudies-literature
Project description:Background and purpose - Classification of ankle fractures is crucial for guiding treatment but advanced classifications such as the AO Foundation/Orthopedic Trauma Association (AO/OTA) are often too complex for human observers to learn and use. We have therefore investigated whether an automated algorithm that uses deep learning can learn to classify radiographs according to the new AO/OTA 2018 standards.Method - We trained a neural network based on the ResNet architecture on 4,941 radiographic ankle examinations. All images were classified according to the AO/OTA 2018 classification. A senior orthopedic surgeon (MG) then re-evaluated all images with fractures. We evaluated the network against a test set of 400 patients reviewed by 2 expert observers (MG, AS) independently.Results - In the training dataset, about half of the examinations contained fractures. The majority of the fractures were malleolar, of which the type B injuries represented almost 60% of the cases. Average area under the area under the receiver operating characteristic curve (AUC) was 0.90 (95% CI 0.82-0.94) for correctly classifying AO/OTA class where the most common major fractures, the malleolar type B fractures, reached an AUC of 0.93 (CI 0.90-0.95). The poorest performing type was malleolar A fractures, which included avulsions of the fibular tip.Interpretation - We found that a neural network could attain the required performance to aid with a detailed ankle fracture classification. This approach could be scaled up to other body parts. As the type of fracture is an important part of orthopedic decision-making, this is an important step toward computer-assisted decision-making.
| S-EPMC7919919 | biostudies-literature
Project description:Objective To evaluate the inter- and intraobserver reliability and reproducibility of the new AO/OTA 2018 classification for distal radius fractures and to compare it with the Fernandez classification system. Method A questionnaire was applied in the Qualtrics software on 10 specialists in hand surgery who classified 50 radiographs of distal radius fractures according to the Fernandez and AO/OTA 2018 classifications and, subsequently, indicated their treatment. The questionnaire was applied in time T0 and repeated after 4 weeks (t1) . The mean agreement between the answers, and the reliability and inter- and intraobserver reproducibility were analyzed using kappa indexes. Results The mean interobserver agreement in the Fernandez classification was 76.4, and it was 59.2% in the AO/OTA 2018 classification. The intraobserver agreements were 77.3 and 56.6%, respectively. The inter- and intraobserver kappa indexes for the Fernandez classification were 0.57 and 0.55, respectively, and, in the AO/OTA 2018 classification, they were 0.34 and 0.31, respectively. Conclusion The AO/OTA 2018 classification showed a low intra- and interobserver reproducibility when compared with the Fernandez classification. However, both classifications have low intra- and interobserver indexes. Although the Fernandez classification did not obtain excellent results, it remains with better agreement for routine use.
| S-EPMC9757955 | biostudies-literature
Project description:BackgroundMulti-fragmentary patella fractures (MFPFs) are common patella fracture type. Low-profile plate fixation seems to be promising advancement in the treatment of such difficult fractures. There is no systematic morphologic study specifically for MFPFs to provide objective reference for the improvement of future implants and biomechanical models. This study aimed to delineate and quantify the location and spatial frequency of fracture lines, comminution zones, and coronal plane fragments in MFPFs using three-dimensional (3D) CT mapping technique.MethodsA total of 187 MFPFs were retrospectively reviewed and analyzed. Fractures were digitally reconstructed from CT data, and fracture lines, comminution zones, and coronal fragments were graphically overlaid onto a 3D patella template. Fracture characteristics were summarized qualitatively based on the fracture maps and quantitatively on the counts and volume of each fragment. Furthermore, according to the classic fracture patterns concerning MFPFs, subgroup analysis was conducted.ResultsOn average, we observed 7 fragments in each fracture, 3 of which were <1 cm3. Most fractures (81.2%) had coronal fragments on the anterior and/or posterior patella surfaces. We identified three classic patella fracture patterns: transverse with comminution, stellate, and "displaced comminuted" in 104, 54, and 29 knees, respectively. 3D maps demonstrated distinct distribution fracture patterns of fracture lines, comminution zones, and coronal fragments.ConclusionsSupero-medial corner of the patella was seldomly involved, and might be used as the cornerstone for fixation. Coronal fragments were common on both anterior and posterior patella surfaces, justifying the application of anterior plate osteosynthesis characterized by multi-planar fixation. Comminution areas mainly concentrated in the lower half of the patella, potentially suitable for an implant in combination with sutures or circumferential cerclage wiring. The described 3D features of MFPFs could provide reference for the design of future implants and biomechanical models.
| S-EPMC8506535 | biostudies-literature
Project description:BackgroundThe reporting and interpretation of data from clinical trials of proximal humeral fractures are hampered by the use of two partly incommensurable fracture classification systems: the Neer classification and the AO/OTA classification. It remains difficult to interpret and generalize results, to conduct prognostic studies, and to obtain consensus on treatment recommendations when concise definitions and a common 'fracture language' are lacking. Thus, we compared both classifications systems using primary data from large clinical studies to assess how thoroughly both systems conveyed clinically important classification information.MethodsClassification data from each study were organized in a cross-table covering the 432 theoretically possible combinations between the 16 Neer categories and the 27 AO/OTA subgroups, and the plausibility of all observed combinations were assessed and discussed by the authors until consensus.ResultsWe analyzed primary data from 2530 observations from seven studies providing primary data from both classification systems. Thirty-five percent (151 out of 432) of the combinations were considered 'not plausible' and thirty-four percent (149 out of 432) were considered 'problematic'.ConclusionsClinically important information was lost within both classification systems. Most important, the varus/valgus distinction was not found within the Neer classification and a clear definition of displacement was lacking in the AO/OTA classification. We encourage surgeons and researches to report data from both classification systems for a more thorough description of the fracture patterns and to enable cross-checking of the coding. A suitable table for cross-checking of the coding is provided herein.
| S-EPMC3610277 | biostudies-literature
Project description:Background and purpose - To achieve a common understanding when dealing with long bone fractures in children, the AO Pediatric Comprehensive Classification of Long Bone Fractures (AO PCCF) was introduced in 2007. As part of its final validation, we present the most relevant fracture patterns in the upper extremities of a representative population of children classified according to the PCCF. Patients and methods - We included children and adolescents (0-17 years old) diagnosed with 1 or more long bone fractures between January 2009 and December 2011 at the university hospitals in Bern and Lausanne (Switzerland). Patient charts were retrospectively reviewed and fractures were classified from standard radiographs. Results - Of 2,292 upper extremity fractures in 2,203 children and adolescents, 26% involved the humerus and 74% involved the forearm. In the humerus, 61%, and in the forearm, 80% of single distal fractures involved the metaphysis. In adolescents, single humerus fractures were more often epiphyseal and diaphyseal fractures, and among adolescents radius fractures were more often epiphyseal fractures than in other age groups. 47% of combined forearm fractures were distal metaphyseal fractures. Only 0.7% of fractures could not be classified within 1 of the child-specific fracture patterns. Of the single epiphyseal fractures, 49% were Salter-Harris type-II (SH II) fractures; of these, 94% occurred in schoolchildren and adolescents. Of the metaphyseal fractures, 58% showed an incomplete fracture pattern. 89% of incomplete fractures affected the distal radius. Of the diaphyseal fractures, 32% were greenstick fractures. 24 Monteggia fractures occurred in pre-school children and schoolchildren, and 2 occurred in adolescents. Interpretation - The pattern of pediatric fractures in the upper extremity can be comprehensively described according to the PCCF. Prospective clinical studies are needed to determine its clinical relevance for treatment decisions and prognostication of outcome.
| S-EPMC5385104 | biostudies-literature
Project description:Background and purpose - To achieve a common understanding when dealing with long bone fractures in children, the AO Pediatric Comprehensive Classification of Long Bone Fractures (AO PCCF) was introduced in 2007. As part of its final validation, we present the most relevant fracture patterns in the lower extremities of a representative population of children classified according to the PCCF. Patients and methods - We included patients up to the age of 17 who were diagnosed with 1 or more long bone fractures between January 2009 and December 2011 at either of 2 tertiary care university hospitals in Switzerland. Patient charts were retrospectively reviewed. Results - More lower extremity fractures occurred in boys (62%, n = 341). Of 548 fractured long bones in the lower extremity, 25% involved the femur and 75% the lower leg. The older the patients, the more combined fractures of the tibia and fibula were sustained (adolescents: 50%, 61 of 123). Salter-Harris (SH) fracture patterns represented 66% of single epiphyseal fractures (83 of 126). Overall, 74 of the 83 SH patterns occurred in the distal epiphysis. Of all the metaphyseal fractures, 74 of 79 were classified as incomplete or complete. Complete oblique spiral fractures accounted for 57% of diaphyseal fractures (120 of 211). Of all fractures, 7% (40 of 548) were classified in the category "other", including 29 fractures that were identified as toddler's fractures. 5 combined lower leg fractures were reported in the proximal metaphysis, 40 in the diaphysis, 26 in the distal metaphysis, and 8 in the distal epiphysis. Interpretation - The PCCF allows classification of lower extremity fracture patterns in the clinical setting. Re-introduction of a specific code for toddler's fractures in the PCCF should be considered.
| S-EPMC5385105 | biostudies-literature
Project description:BackgroundDistal femur fractures are complex injuries with a high rate of fracture healing problems. Since the widespread of computed tomographic imaging in the diagnosis of distal femur fractures, many fracture characteristics have been discovered. This study aimed to depict the location and frequency of distal femur fracture lines and further analyze the morphological characteristics using the 3-dimensional computed tomography (CT) mapping technique, thus providing more information to solve this challenging clinical problem.MethodsIn total, 217 distal femur fractures in 216 patients were retrospectively reviewed. Fracture fragments on CT were digitally reconstructed and virtually reduced to match a template model. The contour of every fracture fragment was then marked with smooth curves, and the overlap of all fracture lines allowed for the creation of 3-dimensional fracture maps and heat maps. Fracture characteristics were summarized based on these maps.ResultsThis study included 114 left knee injuries, 101 right knee injuries, and 1 case with bilateral injury. Distal femur fractures were most likely to occur among patients aged 61 to 70 years. On the heat map of all 217 fractures, fracture line hot zones were mainly concentrated around the metaphysis, the lateral part of the intercondylar notch, and the patellofemoral joint. Distal femur fractures with three Arbeitsgemeinschaft für Osteosynthesefragen Foundation/Orthopaedic Trauma Association (AO/OTA) types demonstrated distinct fracture characteristics. In total, there were 58 coronal plane fractures (41.1%) in 141 intercondylar fractures.ConclusionsThe intercondylar fracture patterns in AO/OTA type B and type C fractures were similar, while the supracondylar characteristics in AO/OTA type A and type C were different. The findings in this study can help orthopaedic surgeons better understand the fracture morphology on the basis of AO/OTA classification. Further studies are needed to establish a standard biomechanical fracture model and new fixation strategy for better clinical outcomes.
| S-EPMC9073806 | biostudies-literature
Project description:AimsDespite limited clinical scientific backing, an additional trochanteric stabilizing plate (TSP) has been advocated when treating unstable trochanteric fractures with a sliding hip screw (SHS). We aimed to explore whether the TSP would result in less post operative fracture motion, compared to SHS alone.MethodsOverall, 31 patients with AO/OTA 31-A2 trochanteric fractures were randomized to either a SHS alone or a SHS with an additional TSP. To compare postoperative fracture motion, radiostereometric analysis (RSA) was performed before and after weightbearing, and then at four, eight, 12, 26, and 52 weeks. With the "after weightbearing" images as baseline, we calculated translations and rotations, including shortening and medialization of the femoral shaft.ResultsSimilar migration profiles were observed in all directions during the course of healing. At one year, eight patients in the SHS group and 12 patients in the TSP group were available for analysis, finding a clinically non-relevant, and statistically non-significant, difference in total translation of 1 mm (95% confidence interval -4.7 to 2.9) in favour of the TSP group. In line with the migration data, no significant differences in clinical outcomes were found.ConclusionThe TSP did not influence the course of healing or postoperative fracture motion compared to SHS alone. Based on our results, routine use of the TSP in AO/OTA 31-A2 trochanteric fractures cannot be recommended. The TSP has been shown, in biomechanical studies, to increase stability in sliding hip screw constructs in both unstable and intermediate stable trochanteric fractures, but the clinical evidence is limited. This study showed no advantage of the TSP in unstable (AO 31-A2) fractures in elderly patients when fracture movement was evaluated with radiostereometric analysis.
| S-EPMC10797560 | biostudies-literature
Project description:BackgroundA lateral approach with open reduction and internal fixation with a plate is a very effective technique for the majority of distal fibular fractures. However, this open approach for ankle fixation may be complicated by wound dehiscence and infection, especially in high-risk patients. An alternative to plating is an intramedullary implant, which allows maintenance of length, alignment, and rotation and which allows for decreased soft tissue dissection. While there has been clinical data suggesting favorable short-term outcomes with these implants, there is no current biomechanical literature investigating this technology in this particular fracture pattern. This study sought to biomechanically compare an emerging technology with an established method of fixation for distal fibular fractures that traditionally require an extensive exposure.MethodsTen matched cadaveric pairs from the proximal tibia to the foot were prepared to simulate an Arbeitsgemeinschaft für Osteosynthesefragen/Orthopaedic Trauma Association (AO/OTA) 44C2 ankle fracture and randomized to fixation with a distal fibular locking plate or intramedullary fibular rod. A constant 700-N axial load was applied, and all specimens underwent testing for external rotation stiffness, external rotation cyclic loading, and torque to failure. The syndesmotic diastasis, stiffness, torque to failure, angle at failure, and mode of failure were obtained from each specimen.ResultsThere was no significant difference in syndesmotic diastasis during cyclic loading or at maximal external rotation between the rod and plate groups. Post-cycle external rotation stiffness across the syndesmosis was significantly higher for the locking plate than the fibular rod. There was no significant difference between the rod and plate in torque at failure or external rotation angle. The majority of specimens had failure at the syndesmotic screw.ConclusionsIn the present cadaveric study of an AO/OTA 44C2 ankle fracture, a modern fibular rod demonstrated less external rotation stiffness while maintaining the syndesmotic diastasis to within acceptable tolerances and having similar failure characteristics.
| S-EPMC5024498 | biostudies-literature