Deep learning-based automated measurement of hip key angles and auxiliary diagnosis of developmental dysplasia of the hip

doi:10.21203/rs.3.rs-4351825/v1

Download PDF

Research Article

Deep learning-based automated measurement of hip key angles and auxiliary diagnosis of developmental dysplasia of the hip

https://doi.org/10.21203/rs.3.rs-4351825/v1

This work is licensed under a CC BY 4.0 License

You are reading this latest preprint version

objectives Radiographic examination is currently the most commonly used method for diagnosing developmental dysplasia of the hip (DDH). In recent years, artificial intelligence (AI) has made significant advances in image recognition, segmentation, decision-making, and statistical analysis of a large number of data sets. Our study aim is whether AI model can be accurately measured angles in pelvic radiography of hip. Then evaluation of AI model effectiveness of pelvic radiographs in diagnosis of DDH and BDDH.

Methods A total of 1029 patients, comprising 273 men and 757 women (aged 18–84 years, median age: 33 years) who underwent pelvic radiography examination between January 2020 and January 2022 were retrospectively included in this study. The images were randomly allocated into the training set (720 cases), validation set (103 cases), and test set (206 cases). The anatomical key points were marked: L-fhc, L-uar, L-tar, L-lt, R-fhc, R-uar, R-tar, and R-lt. The Sharp, Tönnis, and Center edge (CE) angles were calculated automatically based on the above coordinates and corresponding rules. Hip development situation was compared among measurements obtained by the AI model and those obtained manually by two radiologists. The area under the receiver operating characteristic (ROC) curve was used to evaluate the diagnostic effectiveness of the AI model.

Results Manually- and AI model-measured results showed no significant differences in terms of Sharp, Tönnis and Center edge (CE) angles (all P > 0.05). ICCs and correlation coefficient r values were greater than 0.75, indicating that AI model and manual measurements had good repeatability and were positively correlated. AI model measurement results are highly consistent with manual measurement results, with smaller errors. Both AI model and manual measurement results had similar repeatability. The AI model measurement was therefore faster than the radiologists (P < 0.001). AI model measurement had a high diagnostic accuracy, sensitivity and specificity of DDH. AI model has high diagnostic performance for DDH. AI model and manual measurements were basically consistent with clinical diagnosis results (P < 0.05). AI model can be used to evaluate the hip condition by measuring hip sharp, Tönnis and CE angles, which are similar to the clinical diagnosis results and can be used for the auxiliary diagnosis of DDH and BDDH.

Conclusion AI model measurement results are highly consistent with manual measurement results. The AI model measurement was far faster than the radiologists. Sharp, Center edge, and Tönnis angles measured using the deep learning based convolutional neural network model can be used to diagnose DDH and BDDH with a high diagnostic performance. AI model can completely replace manual measurement key angles of hip and diagnosing DDH and BDDH, faster and more precise.

Artificial intelligence

Deep learning

Developmental dysplasia of the hip

Pelvic radiograph

Developmental dysplasia of the hip (DDH) is one of the most common bone and joint diseases encountered in orthopedic practice [1]. DDH encompasses a multitude of pathologies involving the acetabulum, and the proximal femur, such as acetabular dysplasia, hip subluxation, true dislocation of the hip, and hip instability [2]. Another, borderline acetabular dysplasia (BDDH) refers to mildly sub-normal patterns of acetabular shape and coverage that might predispose children to mechanical dysfunction and instability [3]. DDH eventually leads to the development of secondary osteoarthritis if left untreated [4]. In the population screened for DDH, it has an incidence of 5–30% [5]. Radiographic examination is currently the most commonly used method for diagnosing DDH and BDDH [6,3]. However, diagnosis based on measurements obtained from anteroposterior pelvic radiographs is associated with considerable errors and poor reproducibility, for moderate agreement using subjective measures such as the Shenton's line and the outside boundary of the acetabulum is poor [7,8]. The best opportunity for early diagnosis and treatment is therefore missed in many of these patients [9]. In this context, accurate measurement of the hip angle is a major issue that affects early clinical diagnosis and treatment decision-making under this condition [10]. In recent years, artificial intelligence (AI) has made significant advances in image recognition, segmentation, decision-making, and statistical analysis of a large number of data sets [11,12]; it even offers advantages over clinicians in some fields [13]. AI technology, as represented by deep learning, has also been applied to the auxiliary management of diseases related to the hip joint [14]. Fengyu. et al. [15] used a signal-shot multi-box detector to realize hip joint detection on anteroposterior pelvic radiographs. However, they only located the position of the hip joint and did not provide any evidence for the diagnosis of DDH. In another study [16], it determined the anatomical positions of multiple key points by considering the connected edges as anatomical constraints; this was based on the left-right symmetry of 11 pairs of key points. Studies results can be used as reference indicators for the diagnosis of various hip diseases. However, owing to the considerably broad research objectives, the findings cannot provide targeted clinical guidance or facilitate decision-making for a specific disease. We therefore asked: (1) How to more accurately and faster measure the angles of the hip to diagnose DDH and borderline developmental dysplasia of the hip (BDDH)? (2) How consistent and repeatable are manual and AI model measured angles on anteroposterior hip pelvic radiographs? (3) What was evaluation of AI model effectiveness of pelvic radiographs in diagnosis of DDH and BDDH?

Data sets

A total of 1029 patients, comprising 273 men and 757 women (aged 18–84 years, median age: 33 years) who underwent pelvic radiography examination at the Fourth Medical Center of the General Hospital of the Chinese People's Liberation Army between January 2020 and January 2022 were retrospectively included in this study. The original anteroposterior pelvic radiography images were downloaded, and the data were processed by the picture archiving and communication system (PACS). The study is taking clinical diagnosis criteria (clinical presentation, clinical examination, radiographs and ultrasound screening) as the gold standard [1,5].

Only the cases with normal hip joints or with minimal lesions and having clear anatomical key points on the images were included in this study. Cases fulfilling the following criteria were excluded: (1) a history of flat hip deformity or pelvic surgery; (2) hip joint abnormality caused by rheumatism, infection, and tumors; (3) dislocation of the hip joint secondary to central nerve trauma; (4) radiographs demonstrating an indistinguishable pelvic teardrop and the presence of acetabular hyperplasia; and (5) poor posture during radiograph acquisition or lack of imaging data.

The anteroposterior pelvic radiographs obtained in the supine and standing positions were used to constitute an image database. The images were randomly allocated into the training set (720 cases, accounting for 70% of the total), validation set (103 cases, accounting for 10% of the total), and test set (206 cases, accounting for 20% of the total). The training set was used to construct the AI model of hip joint anatomical location key points and the validation set was used to optimize the model parameters. The test set was used to evaluate and compare the key point location, angle measurement, and diagnostic performance for DDH (Fig. 1).

Figure 1 Patients grouping and flow chart.

Instrument and X-ray radiography methods.

The Definium 6000 digital radiography system (GE medical systems, MI, USA) was used to be X-ray the hip examination. The X-ray examination was performed as follows: (1) In the standard supine position: the patient was asked to lie on the back with the tips of both first toes turned inward by 15° and touching each other; the lower limbs were to be kept straight and shoulder-width apart. The line passing through the midpoint of the line between the upper edge of the pubic symphysis and bilateral anterior superior iliac spines was considered as the central line. (2) In the standard standing position: the patient was asked to stand in front of the camera with legs straight and both feet turned slightly inward by 15°. The central line was constructed through a point positioned 2–3 cm above the pubic symphysis.

Measurement parameters and joint marking

The most common and highly recognized hip joint measurement parameters, namely, the Sharp, CE, and Tönnis angles [17,18], were selected (Fig. 2). These parameters were labeled separately by radiologists with 5 years (Dr.1) and 3 years (Dr.2) of related work experience. The marked radiographs were included in the training set after initial marking and re-checking. The following points were marked: the center point of the left femoral head (L-fhc), point on the outer edge of the left acetabulum (L-uar), point on the inner edge of the left acetabulum (L-tar), point on the lower edge of the left tear drop (L-lt), center point of the right femoral head (R-fhc), point on the outer edge of the right acetabulum (R-uar), point on the inner edge of the right acetabulum (R-tar), and point on the lower edge of the right tear drop (R-lt) (Fig. 3). The marked points with differences exceeding 5 mm were re-marked and reviewed by the team to ensure objectivity of the marking.

Figure 2 Schematic diagram of measurement of the CE, Tönnis and Sharp angles

Figure 3 Schematic diagram of hip joint marking

Hip angle measurement and auxiliary diagnosis of DDH and BDDH

The Sharp, Tönnis, and CE angles were calculated automatically based on the above coordinates and corresponding rules. Cases were diagnosed with DDH or borderline developmental dysplasia of the hip (BDDH) based on the following conditions [19]: (1) DDH was diagnosed if the Sharp angle measured > 45 °, (2) DDH was diagnosed if the Tönnis angle measured > 10 °, (3) Borderline developmental dysplasia of the hip (BDDH) was diagnosed if the CE angle measured 20–25°[20,21], while DDH was diagnosed if CE angle measured < 20°.

Data preprocessing

To expand the data size, we collected data obtained by different acquisition devices from the training, validation and test sets, and eliminate differences by preprocessing separately to improve the robustness of subsequent models. The pixel space of all radiographs was unified to 0.2 mm × 0.2 mm using data resampling technology. All pixel gray values were then normalized to the same data range using min-max normalization. Finally, as shown in Fig. 4, all the included images were clipped along the geometric axis based on the symmetry of the anatomical structure of the hip. The left-sided images were flipped to form the input data set for location of the key points of the hip anatomy. This reduced the number of targets for location of hip key points from 8 to 4; in addition, the data sample size expanded and reduced the difficulty of the task. In the training phase, the input radiographs were rotated at random by − 5 ° to 5 ° to further improve the robustness of the model.

Figure 4 Schematic diagram of data preprocessing

Model construction

In this study, we used VB-Net (Visual Basic net) as the basic network for location of the key points of hip anatomical position. Its structure is shown in Fig. 5. The VB-Net replaces the convolutional layer in the traditional U-net network with a bottle neck structure. It greatly compresses the network parameters while maintaining the model performance and substantially improves model training efficiency. During model training, an area of 20 pixels ×20 pixels was selected around key points of each group of anatomical positions as the location target area of VB-Net. In the test phase, the center of gravity of the largest connectivity area with a probability threshold of 0.5 on the probability maps was selected as the prediction output. Combined with spatial information, the coordinates of prediction points of the left-sided image were mapped to the original image coordinate system. The Sharp, Tönnis, and CE angles of the left and right hips were calculated in the test set based on the above definitions.

Figure 5 Sketch map of automated measurement of hip key points based on the VB-Net model

In this study, focal loss was selected as the loss of function of the network; the learning rate was set to 0.0001. Data preprocessing and automated parameter calculation were performed using Python 3.8. The training and testing of the deep learning model were performed using the PyTorch framework on the NVIDIA GeForce GTX 1080Ti graphics card.

AI Model effectiveness evaluation and statistical analysis

Two radiologists (Doctor.1, Doctor.2) measured the Sharp, Tönnis, and CE angles in the test set using the measuring rule provided by the picture archiving and communication system (PACS). Due to the lack of a unified measurement reference standard, the average value of each measurement index identified by the two radiologists (Dr.1, Dr.2) was used as the reference standard. Hip development situation was compared among measurements obtained by the AI model and those obtained manually by radiologists.

Data analysis and sorting were performed using SPSS software version 25.0 (Armonk, NY: IBM Corp). Measurement data were expressed as the mean ± standard deviation. Comparisons among multiple groups were performed using the F test for variances and groups comparisons (Dr.1, Dr.2 and AI model) were performed using the least significant difference test. The correlation between measurements obtained by the model and those obtained manually by radiologists (Dr.1, Dr.2) was analyzed using Pearson’s correlation. The intra-class correlation coefficient (ICC) was used to evaluate the consistency of measurement results; ICC ≥ 0.75 indicated good consistency. The area under the receiver operating characteristic (ROC) curve was used to evaluate the diagnostic effectiveness of the AI model; P < 0.05 was considered statistically significant.

Evaluation of the accuracy of AI measurement results

Model and manual measurements of Sharp, Tönnis and CE angles from 206 anteroposterior pelvic radiographs of the test set were compared. The accuracy of model measurement results was evaluated and related results are shown in Table 1.

Table 1

Difference analysis between manual and model measurements (test set, n = 206)
		Manual measurement		AI Model measurement	F	P
		Dr. 1	Dr. 2	AI Model measurement	F	P
Left	Sharp angle	43.4°±5.2°	43.2°±5.3°	43.2°±5.9°	0.03	0.99
	Tönnis angle	12.2°±7.8°	11.9°±7.8°	12.3°±9.7°	0.08	0.97
	Center edge angle (CE)	25.4°±10.1°	25.4°±10.2°	25.6°±9.6°	0.05	0.98
Right	Sharp angle	43.5°±5.6°	43.1°±5.5°	43.0°±5.4°	0.36	0.78
	Tönnis angle	11.6°±8.5°	11.1°±8.1°	11.8°±9.5°	0.30	0.83
	Center edge angle (CE)	25.0°±10.2°	24.7°±10.4°	26.5°±9.9°	1.36	0.26
Note: Dr., doctor; AI, artificial intelligence.

Manually- and AI model-measured results showed no significant differences in terms of Sharp, Tönnis and CE angles (all P > 0.05).

Repeatability and ICC values for Sharp, Tönnis and CE angles measured by the AI model and those measured manually were analyzed, and results are shown in Fig. 6. The ICC values for the left Sharp, Tönnis and CE angles were 0.801, 0.757 and 0.895, respectively, with the corresponding r values of 0.807, 0.774 and 0.896, respectively. The ICC values for the right Sharp, Tönnis and CE angles were 0.943, 0.790 and 0.908, respectively, with the corresponding r values of 0.946, 0.794, and 0.921, respectively. In the above results, ICCs and correlation coefficient r values were greater than 0.75, indicating that model and manual measurements had good repeatability and were positively correlated.

Figure 6 Bland-Altman plot showing the consistency of artificial intelligence (AI) model measurement and manual measurement (MM) at various angles.

It can be seen from Fig. 6 that all scatter points are distributed within the standard deviation line, and the mean line is close to 0, indicating that the results of the two measurement methods (manual measurement and AI model measurement) are relatively close and have good consistency. Further analysis showed that the mean absolute errors of model measurement of Sharp, Tönnis and CE angles were 1.33°, 2.30° and 2.68° on the left side, and 1.12°, 2.35° and 3.05° on the right side, respectively. The root means square errors of model measurement of Sharp, Tönnis and CE angles were 3.49°, 6.13° and 4.48° respectively on the left side and they were 1.85°, 5.86° and 4.36 ° respectively on the right side. These results suggest that AI model measurement results are highly consistent with manual measurement results, with smaller errors. Both AI model and manual measurement results had similar repeatability.

Comparison of time taken for hip measurement

Measurement results from 206 images showed no significant difference between the AI model and two radiologists (Dr.1, Dr.2) in terms of the time (second) taken for hip measurement (Fig. 7).

Figure 7 Average time taken by the artificial intelligence (AI) model and two radiologists for performing measurements on each image

Radiologists took longer time for performing the hip measurements compared with AI model; and AI model took successively shorter times. The AI model measurement was therefore faster than the radiologists (P < 0.001).

Evaluation of DDH diagnosis performance by model measurement

The results of AI automated measurement of the hip based on anteroposterior pelvic radiographs were analyzed. The three parameters used to evaluate hip development were used for the auxiliary diagnosis of DDH (Table 2).

Table 2

DDH diagnosis based on artificial intelligence model measurement
Hip	Angle	Accuracy	Sensitivity	Specificity
Left	Sharp	89.8%	82.1%	94.5%
	Tönnis	90.1%	90.3%	91.4%
	Center edge (CE)	86.8%	84.8%	87.9%
Right	Sharp	93.7%	88.6%	96.3%
	Tönnis	92.2%	91.3%	93.2%
	Center edge (CE)	80.5%	90.9%	76.4%

The results showed that all angles by AI model measurement had a high diagnostic accuracy, sensitivity, and specificity of DDH.

Comparison of DDH and BDDH diagnosis by artificial intelligence models and radiologists

The receiver operating characteristic (ROC) curves of AI model and manual measurements were used to evaluate DDH diagnosis (Fig. 8).

Figure 8 ROC curves plotted based on AI model and manual measurements of hip Sharp and Tönnis angles for diagnosis of DDH

It can be seen from the Fig. 8 that the area under curve (AUC) of ROC of AI model measuring Sharp angle (left 0.883, right 0.924) and Tönnis angle (left 0.908, right 0.922) are very close to doctors, which indicates that AI model has high diagnostic performance for DDH.

It can be seen from Fig. 8c that the CE angle measurements by AI model can help diagnose DDH with a high diagnostic performance (AUC of AI model measuring CE angle: left 0.922, right 0.871). At same time, the CE angle measurements by AI model can also help diagnose BDDH with a high diagnostic performance from Fig. 8c (AUC of AI model measurements CE angle: left 0.787, right 0.676).

Comparison of AI model and manual diagnosis results of DDH with clinical diagnosis results

Taking clinical diagnosis results as the gold standard, the Sharp angle (AI model A), Tönnis angle (AI model B), CE angle (AI model C) and manual measurement results were compared with the final clinical diagnosis results. Kappa tests were performed (Tables 3,4,5).

Table 3

Comparison of AI model A (Sharp angle) and manual measurement diagnosis results with clinical diagnosis results (n = 206)
	Diagnosis results	Diagnosis results
		Left				Right
		Normal	DDH	Kappa	P	Normal	DDH	Kappa	P
AI model A	Normal	121	14	0.779	0.000	131	8	0.858	0.000
	DDH	7	64			5	62
Dr.1	Normal	125	7	0.898	0.000	133	4	0.913	0.000
	DDH	3	71			4	65
Dr.2	Normal	121	6	0.860	0.000	133	6	0.890	0.000
	DDH	7	72			4	63
Note: DDH: developmental dysplasia of the hip; Dr: doctor; Normal: normal hip.

Table 4

Comparison of AI model B (Tönnis angle) and manual measurement diagnosis results with clinical diagnosis results (n = 206)
	Diagnosis results	Diagnosis results
		Left				Right
		Normal	DDH	Kappa	P	Normal	DDH	Kappa	P
AI model B	Normal	85	11	0.814	0.000	96	9	0.845	0.000
	DDH	8	102			7	94
Dr.1	Normal	92	1	0.951	0.000	102	5	0.942	0.000
	DDH	4	109			1	98
Dr.2	Normal	87	7	0.873	0.000	98	6	0.893	0.000
	DDH	6	106			5	97
Note: DDH: developmental dysplasia of the hip; Dr: doctor; Normal: normal hip.

Table 5

Comparison of AI model C (CE angle) and manual measurement diagnosis results with clinical diagnosis results (n = 206)
	Diagnosis results	Diagnosis results
		Left					Right
		Normal	BDDH	DDH	Kappa	P	Normal	BDDH	DDH		Kappa	P
AI mode C	Normal	51	5	1	0.778	0.000	42	2		1	0.663	0.000
	BDDH	3	28	7			13	20		2
	DDH	1	11	99			0	23		103
Dr.1	Normal	98	9	0	0.866	0.000	103	6		0	0.873	0.000
	BDDH	5	34	2			3	34		2
	DDH	0	1	57			0	5		53
Dr.2	Normal	98	9	0	0.835	0.000	103	6		1	0.840	0.000
	BDDH	4	32	3			3	34		3
	DDH	1	4	55			2	5		49
Note: DDH: developmental dysplasia of the hip; Dr: doctor; Normal: normal hip; BDDH: borderline developmental dysplasia of the hip.

The results of the kappa test showed that in terms of Sharp, Tönnis and CE angles, AI model and manual measurements were basically consistent with clinical diagnosis results (P < 0.05). This suggests that AI model can be used to evaluate the hip condition by measuring hip sharp, Tönnis and CE angles, which are similar to the clinical diagnosis results and can be used for the auxiliary diagnosis of DDH and BDDH.

The hip, as one of the largest load-bearing joints in the human body, is mainly responsible for providing dynamic support to the weight of the upper body; it also promotes the axial transmission of stress and load to the lower limbs [22]. Its ball and socket structure offers obvious horizontal stability and can allow movement in three planes: flexion/extension, abduction/adduction, and internal/external rotation [23]. Studies [1, 5,24] have shown that persistent dysplasia of the hip may lead to changes in body posture and gait; they also suggest that the limbs of patients with reduced cartilage or other degenerative joint diseases demonstrate compensatory changes in the form of shortening. This results in postural scoliosis, backache, or even disability, which has considerable impact on the quality of life of patients. The Sharp, Tönnis, and CE angles are important indicators for the clinical evaluation of hip joint stability; they can be used to evaluate acetabular development and its coverage of the femoral head [25]. Previous studies have found the range of measurement errors between different observers to be ± 3.5° to ± 10° [8]. Yinjun. et al. [26] used a deep learning algorithm for automated measurement of the Sharp angle of the hip and found the mean absolute error for the left and right sides to be 1.57° and 1.73°, respectively. Our results showed the mean absolute error for the left and right Sharp angles to be 1.33° and 1.12°, respectively; this was lower than those observed by Yinjun. et al. [26]. We also found no significant differences between AI model- and manually measured Sharp, Tennis, and CE angles. However, the AI model-measured Sharp angles were smaller for both the left and right sides.

The inclusion of osteophytes at the outer edge of the acetabulum affects the magnitude of the measured hip angles. The magnitude of the measured CE angle may therefore be significantly increased if osteophytes are not clearly distinguished [25]. As none of the cases included in this study showed any obvious osteophytes at the outer edge of the acetabulum, this factor is not likely to have had a substantial impact on the results. Qiang. et al. [27] had developed a model for automated measurement of the CE angle; the results provided by the model did not differ significantly from those measured by the radiologists. Holden A et al. [28] measured the Sharp, tennis, and CE angles after constructing a model that automatically located hip joint key points. They found good consistency between the manually- and model-measured results. we replicated this finding and demonstrated the practicability of the AI model for DDH and BDDH diagnosis. In the previous studies by Qiang. et al. [27], the constructed models took an average time of 1.19 and 1.22 seconds to automatically measure the Sharp and CE angles, respectively. In our study, the constructed model took an average time of 1.7 seconds to simultaneously measure the Sharp, Tönnis, and CE angles, while two radiologists took an average time of 88.1 ± 8.4 (DR.1) and 90.3 ± 7.4 (DR.2) seconds, respectively, to measure the 3 angles; the model will therefore greatly improve radiologist efficiency for measurement. Compared with Li's model, our model adds the measurement of Tönnis and CE angles, which takes similar time and improved the accuracy of diagnosis by adding test methods and provide more possibilities for clinical application.

Despite there are still controversies about the diagnosis of BDDH [29], the coverage level of the femoral head and acetabulum may be divided into borderline developmental dysplasia (BDDH) and DDH based on various degrees of change, represented by an exceptionally shallow or upwardly inclined acetabulum; these changes can lead to uneven stress on the hip [15]. So, the hips with CE angles of < 20° or 20°–25° (or 18°–24°) were considered to have DDH or BDDH [3,21]. In this study, respectively, for DDH, the diagnostic performance of the CE angle was 0.922 and 0.871 for the left and right sides respectively in AI model; and for BDDH, the diagnostic performance of the CE angle was 0.787 and 0.676 for the left and right sides respectively in AI model. This suggests that measurement using an artificial intelligence-based model is of considerable value in the diagnosis of DDH and BDDH. The hips with a Sharp angle of > 45° and Tönnis angle of > 10° were identified to be dysplastic; the diagnostic accuracy of these values for DDH were 93.7% and 92.2%, respectively, for the left side and 89.8% and 90.1%, respectively, for the right side.

This study has certain limitations. First, manual marking depends on the subjective experience of the observer. As it cannot provide objective standard values, a certain difference exists between AI model and manual measurements in terms of diagnostic accuracy. Second, external validation was not performed in this study; only single data processing was performed at our hospital. There may therefore be limitations to the universality of the model. Although the model developed in this study can measure the Sharp, Tönnis, and center edge angles, a single model can only provide a single indicator for evaluation; numerous challenges are therefore expected to its future application in clinical practice. At the same time, in order to better screen diagnosis and treatment, we need to further study the measurement models of femoral neck stem angle, acetabular coverage and acetabular depth, establish CT and MRI data models, and jointly build a multimodal intelligent DDH diagnosis and treatment system to improve the detection rate of DDH.

In this study, a deep learning algorithm was used to identify the key anatomical points on anteroposterior pelvic radiographs and construct a deep learning algorithm-based convolutional neural network model for automated measurement of Sharp, Center Edge (CE), and Tönnis angles. The effectiveness of the AI model in the auxiliary diagnosis of DDH was also evaluated. AI model measurement results are highly consistent with manual measurement results. And both AI model and manual measurement results had similar repeatability. The AI model measurement was far faster than the radiologists. Sharp, CE, and Tönnis angles measured using the deep learning based convolutional neural network model can be used to diagnose DDH and BDDH with a high diagnostic performance. AI model can completely replace manual measurement key angles of hip and diagnosing DDH and BDDH, faster and more precise.

DDH Developmental Dysplasia of the Hip

BDDH Borderline Acetabular Dysplasia

AI Artificial Intelligence

PACS Picture Archiving and Communication System

L-fhc Center Point of the Left Femoral Head

L-uar Point on the Outer Edge of the Left Acetabulum

L-tar Point on the Inner Edge of the Left Acetabulum

L-lt Point on the Lower Edge of the Left Tear Drop

R-fhc Center Point of the Right Femoral Head

R-uar Point on the Outer Edge of the Right Acetabulum

R-tar Point on the Inner Edge of the Right Acetabulum

R-lt Point on the Lower Edge of the Right Tear Drop

CE Center Edge

VB-Net Visual Basic Net

ROC The Receiver Operating Characteristic Curves

Acknowledgements

We express our gratitude to Department of Research and Development of United Imaging Intelligence(Beijing) Co.Ltd. for their support, for example Xiaoming Liu MD’s works. TianRan Li had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis. All authors have read and approved the final version of manuscript.

Author contributions

RXL, XX, BBZ and TRL wrote the main manuscript text, and RXL, XX,XML prepared all Figs. RXL, XX and WHL participated in the data collection, and analysis. WX and TRL contributed to the study design and conception, and revision of the manuscript. All authors reviewed the manuscript.

Funding

This study was not supported by any research fund.

Data availability

The datasets supporting the conclusions of this article are included within the article.

Data availability

The datasets used and/or analysed during the current study available from the corresponding author on reasonable request.

Ethics approval and consent to participate

The study was approved by the Institutional Review Board of the Fourth Medical Center of the General Hospital of the Chinese People's Liberation Army. All methods are conducted in accordance with relevant guidelines and regulations. All methods in the study were carried out in accordance with the Helsinki guidelines and declaration. We confirm that informed consent was obtained from all patients. Each author certifies that all investigations were conducted in conformity with ethical principles.

Consent for publication

Not Applicable.

Competing interests

The authors declare no competing interests.

Author details

1 Department of Radiology, The Fourth Medical Center of Chinese PLA General Hospital, Beijing, China.

2 Department of Research and Development, United Imaging Intelligence(Beijing) Co.Ltd., Beijing, China.

Sioutis S, Kolovos S, Papakonstantinou ME, et al. Developmental dysplasia of the hip: a review. J Long Term Eff Med Implants. 2022;32(3):39–56.
Anand SD, Muzammil A, Daniel R, et al. Current Surgical Techniques in the Treatment of Adult Developmental Dysplasia of the Hip. J Pers Med. 2023;13(6): 942.
Sarah. DB, Michael. BM. The borderline dysplastic hip: when and how is it abnormal? Pediatr Radiol. 2019;49(12):1669–1677.
Bridget KE, Ernest LS, Shevaun MD. Adolescent hip dysplasia: what are the symptoms and how to diagnose it. Curr Opin Pediatr. 2021;33(1):65–73.
Zhang S, Doudoulakis KJ, Khurwal A, et al. Developmental dysplasia of the hip. Br J Hosp Med. 2020;81(7):1–8.
Schmitz MR, Murtha AS, Clohisy JC, et al. Developmental Dysplasia of the Hip in Adolescents and Young Adults. J Am Acad Orthop Surg. 2020;28(3):91–101.
Weize X, Liqi S, Ping G, et al. A deep learning aided diagnostic system in assessing developmental dysplasia of the hip on pediatric pelvic radiographs. Front Pediatr. 2022; 8:9:785480.
Sicheng Z, Jun S, Chuanbin L, et al. Clinical application of artificial intelligence-assisted diagnosis using anteroposterior pelvic radiographs in children with developmental dysplasia of the hip. Bone Joint J. 2020;102-B(11):1574–1581.
Sheridan GA, Nagle M, Howells C, et al. A radiographic clinic for developmental dysplasia of the hip (DDH). Ir J Med Sci. 2020;189(1):27–31.
Guangyao Y, Yaoxian J, Tong L, et al. A semi-automatic diagnosis of hip dysplasia on x-ray films. Front Mol Biosci. 2020; 17:7:613878.
Qiang L, Lei Z, Hongnian H, et al. Auxiliary diagnosis of developmental dysplasia of the hip by automated detection of Sharp's angle on standardized anteroposterior pelvic radiographs. Medicine (Baltimore). 2019; 98(52): e18500.
Long W, XiaoChun Y, Jiang W,et al. Short-term outcome of artificial intelligence-assisted preoperative three-dimensionalplanning of total hip arthroplasty for developmental dysplasia of the hip compared to traditional surgery. Jt Dis Relat Surg. 2023,34(3):571–582.
Szoldán P, Egyed Z, Szabó E, et al. Segmentation of knee cartilages in MR images with artificial intelligence. Orv Hetil. 2021;162(9):352–360.
Hasan S, Ahmed A, Waheed MA, et al.Transforming Orthopedic Joint Surgeries: The Role of Artificial Intelligence (AI) and Robotics. Cureus. 2023;15(8):e43289.
Fengyu L, Chihchi C, Chitung C, et al. Automatic hip detection in anteroposterior pelvic radiographs a label less practical framework. J Pers Med. 2021;11(6):522.
Wei L, Yu W, Tao J, et al. Landmarks Detection with Anatomical Constraints for Total Hip Arthroplasty Preoperative Measurements. International Conference on Medical Image Computing and Computer Assisted Intervention.2020:670–679. Available at: https://link.springer.com/chapter/10.1007/978-3-030-59719-1_65.
Wei Y, Qin Y, Shuai M, et al. Feasibility of automatic measurements of hip joints based on pelvic radiography and a deep learning algorithm. Eur J Radiol. 2020;132(11):109303.
Ameya B, Mukai CG, Martin B, et al. The variation in hip stability measurements between supine and standing radiographs of dysplastic hips. Bone Joint J. 2021;103-B(11):1662–1668.
Harris JD, Lewis BD, Park KJ. Hip Dysplasia. Clin Sports Med. 2021;40(2):271–288.
Scialpi M, Bezza M, Zottele F, et al. Echography in the study of the neonatal hip: the diagnostic-therapeutic implications of type-IIa (borderline) findings. Radiol Med.1992;83(4):419–422.
Jia Z, Chunbao L, Mingyang A, et al. Chinese expert consensus on the diagnosis and treatment of borderline developmental dysplasia of the hip (2022 Edition). Chin Med J (Engl). 2023; 136(9):1012–1014.
Polkowski GG, Clohisy JC. Hip biomechanics. Sports Med Arthrosc Rev. 2010;18(2):56–62.
Flack NA, Nicholson HD, Woodley SJ. A review of the anatomy of the hip abductor muscles, gluteus medius, gluteus minimus, and tensor fascia lata. Clin Anat. 2012;25(6):697–708.
Scott Y, Natalie Z, Elizabeth L, et al. Developmental Dysplasia of the Hip. Pediatrics. 2019;143(1): e20181147.
Yaoxian J, Guangyao Y, Yuan L, et al. Computer-Aided System Application Value for Assessing Hip Development. Front Physiol. 2020; 1:11:587161.
Yinjun L, Guangyao Y, Yaoxian J, et al. Measuring lateral center edge angle and Sharp angle of hip joint on X-ray films based on deep learning for diagnosing hip dysplasia. Chin J Med Imaging Technol(Chinese). 2022;38(11):1710–1714.
Qiang L, Wenzhuo Y, Meng X, et al. Model construction and application for automated measurement of CE angle on pelvis orthograph based on MASK-R-CNN algorithm. Biomed Phys Eng Express. 2021;7(3).
Holden A, Seth R, Ahmed A, et al. Artificial intelligence-generated hip radiological measurements are fast and adequate for reliable assessment of hip dysplasia: an external validation study. Bone Jt Open. 2022;3(11):877–884.
Jinyan W, Xiaodong C. Recent research progress of hip-preserving treatment for adolescents and adults with developmental dysplasia of the hip. Zhongguo Xiu Fu Chong Jian Wai Ke Za Zhi. 2021,35(12):1513–1518.(Chinese)

No competing interests reported.

Download PDF

Editorial decision: Revision requested
08 Jul, 2024
Reviews received at journal
08 Jul, 2024
Reviewers agreed at journal
07 Jul, 2024
Reviews received at journal
10 Jun, 2024
Reviewers agreed at journal
31 May, 2024
Reviewers invited by journal
31 May, 2024
Editor invited by journal
23 May, 2024
Submission checks completed at journal
19 May, 2024
Editor assigned by journal
19 May, 2024
First submitted to journal
30 Apr, 2024

You are reading this latest preprint version

Deep learning-based automated measurement of hip key angles and auxiliary diagnosis of developmental dysplasia of the hip

Status:

Version 1

Abstract

Figures

Background

Methods

Data sets

Measurement parameters and joint marking

Hip angle measurement and auxiliary diagnosis of DDH and BDDH

Data preprocessing

Model construction

AI Model effectiveness evaluation and statistical analysis

Results

Evaluation of the accuracy of AI measurement results

Comparison of time taken for hip measurement

Evaluation of DDH diagnosis performance by model measurement

Comparison of DDH and BDDH diagnosis by artificial intelligence models and radiologists

Comparison of AI model and manual diagnosis results of DDH with clinical diagnosis results

Discussion

Conclusion

Abbreviations

Declarations

References

Additional Declarations

Status:

Version 1