Radiographic classifications in Perthes disease

An AI based classifier model for lateral pillar classification of Legg-Calve-Perthes

We intended to compare the doctors with a convolutional neural network (CNN) that we had trained using our own unique method for the Lateral Pillar Classification (LPC) of Legg-Calve-Perthes Disease (LCPD). Thousands of training data sets are frequently required for artificial intelligence (AI) applications in medicine. Since we did not have enough real patient radiographs to train a CNN, we devised a novel method to obtain them. We trained the CNN model with the data we created by modifying the normal hip radiographs. No real patient radiographs were ever used during the training phase. We tested the CNN model on 81 hips with LCPD. Firstly, we detected the interobserver reliability of the whole system and then the reliability of CNN alone. Second, the consensus list was used to compare the results of 11 doctors and the CNN model. Percentage agreement and interobserver analysis revealed that CNN had good reliability (ICC = 0.868). CNN has achieved a 76.54% classification performance and outperformed 9 out of 11 doctors. The CNN, which we trained with the aforementioned method, can now provide better results than doctors. In the future, as training data evolves and improves, we anticipate that AI will perform significantly better than physicians.

Predictors of radiographic outcomes of conservative and surgical treatment of Legg-Calvé-Perthes disease

PURPOSE

Treatment outcomes of conservative and surgical treatment of Legg-Calvé-Perthes disease (LCPD) have been shown to be conditioned by a number of factors that may vary across different populations. This retrospective study aimed to evaluate factors affecting radiographically assessed treatment outcomes in patients treated surgically or conservatively for LCPD at Faculty Hospital Motol, Prague, Czech Republic, between the years 2006 and 2019.

METHODS

Data of forty-seven children comprising 52 hips were analysed. Treatment outcomes were evaluated according to Stulberg classification. Predictors included the initial stage of fragmentation of the hip joint according to Herring classification, type of treatment (conservative or surgical), age at the time of diagnosis and sex.

RESULTS

Older age and severity of LCPD according to Herring classification but not the type of treatment were the strongest factors determining treatment outcomes. Treatment outcomes were comparable in patients treated conservatively or surgically both across the whole cohort of patients and a group of young children < six years of age.

CONCLUSIONS

Results strengthen the roles of severity of the LCPD at onset of treatment and age of the patient in predicting treatment outcomes in patients with LCPD. Conservative and surgical treatments appear to yield similar treatment outcomes irrespective of age of patients.

Inter- and intra-rater reliability of the Oberg-Manske-Tonkin classification of congenital upper limb anomalies

On two occasions, five surgeons classified a cohort of 150 consecutive patients with congenital upper limb anomalies according to the Oberg-Manske-Tonkin classification (2020 update). We estimated reliability for the main anomaly code by means of Cohen's kappa (Κ) for ten rater pairs for five common and easily distinguishable anomalies (Group 1), and for all the other anomalies (Group 2). Inter-rater reliability for all patients (n = 150) was substantial, almost perfect for Group 1 (n = 64), but only moderate for Group 2 (n = 86). Intra-rater reliability was higher for all groups. We suggest simplifications to the Oberg-Manske-Tonkin classification and highlight specific requirements for instructions to increase its reliability.Level of evidence: I.

[Acquired hip joint pathologies in childhood]

A number of different acquired pathologies can affect the pediatric hip joint. The classical and relatively common reactive, inflammatory or traumatic entities have to be differentiated from each other. Rarer and very rare pathologies, such as neoplastic entities or tenosynovial giant cell tumor, must also be considered in the differential diagnostics. The correct diagnosis, treatment planning and follow-up monitoring require close coordination between the departments of pediatric radiology and pediatric orthopedics. Sonography often represents the initial diagnostic step, followed by conventional radiography. Further evaluation of hip joint pathologies requires magnetic resonance imaging. In selected cases computed tomography can also be indicated.

Validation of the height-width measurement in Perthes disease among interpreters of different experience levels

The epiphyseal height and metaphyseal width ratio estimated lateral pillar involvement with limited generalizability among various interpreters. The aim of the study was to evaluate the reliability and the validity of height-width ratio measurement in Perthes disease among interpreters of different experience levels. A cross-sectional study was conducted between 2019 and 2020. We included four groups of interpreters: orthopaedic residents, orthopaedic fellows, radiology residents and radiology fellows who were unaware of radiographic study materials. Each interpreter blindedly evaluated Perthes hip radiographs twice at 1-month intervals using the height-width ratio method. Patients' and interpreters' characteristics, height-width ratio and height-width ratio converted to lateral pillar classification (A, B and C) were collected. Intra- and interobserver reliability validated with the paediatric orthopaedist were estimated. Twenty-four interpreters assessed 18 Perthes radiographs. Intraobserver level of agreement (95% confidence interval) for height-width ratio was 0.022 (-0.017 to 0.062), -0.027 (-0.074 to 0.019), -0.010 (-0.095 to 0.074) and 0.019 (-0.109 to 0.146); and interobserver reliability compared with the paediatric orthopaedist was -0.007 (-0.091 to 0.077), 0.003 (-0.056 to 0.061), -0.021 (-0.077 to 0.035) and -0.002 (-0.090 to 0.086) for orthopaedic residents and fellows, radiology residents and fellows, respectively. Kappa statistics of height-width ratio converting to lateral pillar classification indicated intraobserver agreement of orthopaedic residents and fellows, radiology residents and fellows was 0.83, 0.75, 0.54 and 0.91; and interobserver agreement compared with the paediatric orthopaedist was 0.92, 0.83, 0.42 and 0.83, orderly. Height-width ratio and lateral pillar estimation by orthopaedic and radiology trainees have moderate to excellent reliability. Level of evidence: Level II - diagnostic study.

Reliability of the modified lateral pillar classification for Legg Calvé Perthes disease performed by a large group of international paediatric orthopaedic surgeons

PURPOSE

The modified lateral pillar classification (mLPC) is used for prognostication in the fragmentation stage of Legg Calvé Perthes disease. Previous reliability assessments of mLPC range from fair to good agreement when evaluated by a small number of observers with pre-selected radiographs. The purpose of this study was to determine the inter-observer and intra-observer reliability of mLPC performed by a group of international paediatric orthopaedic surgeons. Surgeons self-selected the radiograph for mLPC assessment, as would be done clinically.

METHODS

In total, 40 Perthes cases with serial radiographs were selected. For each case, 26 surgeons independently selected a radiograph and assigned mLPC and 21 raters re-evaluated the same 40 cases to establish intra-observer reliability. Rater performance was determined through surgeon consensus using the mode mLPC as 'gold standard'. Inter-observer and intra-observer reliability data were analysed using weighted kappa statistics.

RESULTS

The weighted kappa for inter-observer correlation for mLPC was 0.64 (95% confidence interval: 0.55 to 0.74) and was 0.82 (range: 0.35 to 0.99) for intra-observer correlation. Individual surgeon's overall performance varied from 48% to 88% agreement. Surgeon mLPC performance was not influenced by years of experience (p = 0.51). Radiograph selection did not influence gold standard assignment of mLPC. There was greater agreement on cases of mild B hips and severe C hips.

CONCLUSIONS

mLPC has low good inter-observer agreement when performed by a large number of surgeons with varied experience. Surgeons frequently chose different radiographs, with no impact on mLPC agreement. Further refinement is needed to help differentiate hips on the border of group B and C.

LEVEL OF EVIDENCE

III.