Efficacy of an artificial intelligence preoperative planning system for assisting in revision surgery after artificial total hip arthroplasty

Abstract

Objective

To explore the early efficacy of an artificial intelligence preoperative planning system (AIHIP system) for assisting in hip revision surgery.

Methods

The clinical data of 25 patients (26 hips) who underwent hip revision between June 2019 and December 2023 and who met the selection criteria were retrospectively analyzed. There were 13 males and 12 females; the ages ranged from 44 to 90 years, with a mean of 69.1 years. The patients' replacement of prosthesis model, operation time, hospitalization time, postoperative time out of bed, etc., as well as the occurrence of adverse events such as postoperative infection, fracture, and loosening of the prosthesis were recorded. The Harris Hip score (HHS) was used to evaluate the function of the affected limbs preoperatively, and 1 week and 6 months postoperatively, and hip mobility was compared preoperatively and 6 months postoperatively.

Results

All 25 patients were followed up for 6 to 59 months, with an average of 25.3 months. Except for one patient who developed a thigh hematoma (treated with incision and drainage and decompression) and hip dislocation in one hip (repaired), the remaining patients experienced no adverse events such as loosening of the prosthesis or infection. The postoperative acetabular cup type matching degree completely matched 25 hips, not matching 1 hip (+ 2 number), for a matching rate of 96.15%; the femoral stem type matching degree completely matched 25 hips, generally matching 1 hip (-1 number), for a matching rate of 100%. The Harris scores were 54.7 ± 9.6 and 89.6 ± 7.0 at 1 week and 6 months after surgery, respectively, which were significantly improved (P < 0.05) compared with the preoperative scores of 33.5 ± 8.3, and further improved at 6 months after surgery compared with the 1-week period (P < 0.05). The patients' hip function was evaluated according to the Harris score at 6 months after surgery, and they were assigned to 23 good hips and 3 medium hips, which could satisfy daily life needs. Hip mobility at 6 months after surgery was 111.15 ± 9.72°, and the difference was statistically significant compared with the preoperative value of 79.42 ± 17.51° (t = -8.077, P < 0.001).

Conclusion

AIHIP system-assisted treatment of THA postoperative revision patients can improve the precision of revision surgery, and reduce the difficulty of surgery, in patients with good postoperative recovery and satisfactory early outcomes.

Introduction

At present, total hip arthroplasty (THA) has been widely used in the clinical treatment of hip dysplasia, femoral head necrosis, hip osteoarthritis and other diseases. With the prolongation of the use of artificial hip joints, the number of patients who need to undergo hip revision surgery is increasing annually due to aseptic loosening of the articular prosthesis, fracture, peripheral osteolysis, infection and even dislocation [1,2,3,4]. Although the success rate of THA surgery is increasing due to the continuous improvement of hip prosthesis materials and structures and surgical techniques, surgical failure due to aseptic loosening of the prosthesis is still one of the main reasons for performing hip revision surgery [5]. In hip revision surgery, the treatment of bone defect and the choice of prosthesis position are the key to the success of the surgery [6], and the failure of THA surgery often leads to a huge bone defect in the acetabulum, which greatly increases the difficulty of hip revision surgery; the intraoperative removal of residual bone cement or the extraction of fixed prosthesis stems is very likely to result in further bone defects or fractures, which will affect postoperative efficacy. Therefore, strict preoperative planning is required before hip revision surgery.

Currently, most hospitals in China still use traditional two-dimensional or three-dimensional preoperative planning software, but there are problems such as the need to manually segment the measurements, time consumption, and other problems, which make it difficult to meet the needs of clinical surgery [7,8,9]. Artificial intelligence (AI) is an emerging technology, and AI has shown initial success in assisting THA surgery [10,11,12]. The clinical data of hip revision surgery patients admitted to the Affiliated Hospital of Nanjing University of Traditional Chinese Medicine who received preoperative planning via the AI preoperative planning system (AIHIP system) from June 2019 to December 2023 were reviewed and analyzed to investigate the clinical value of the system. The report is as follows.

Materials and methods

Patient selection criteria

The inclusion criteria for patients were as follows: ① postoperative failure of THA due to loosening, fracture, infection or even dislocation of the prosthesis; ② progressive bone loss; ③ age ≥ 40 years; ④ preoperative imaging data meeting the selection criteria of the AIHIP system (with hip joint frontal and lateral X-ray and hip CT scanning to the entire pelvis and 15 cm below the femoral tuberosity); and ⑤ follow-up time ≥ 6 months and complete data.

The exclusion criteria were as follows: ① Combined contraindications to surgery such as coagulation disorders, which prevented completion of surgery; ② severe osteoporosis of the hip joint or surrounding, or the presence of neuromuscular dysfunction; ③ the presence of active infectious lesions in the hip joint or in vivo; and ④ failure to undergo surgery after preoperative planning using the AIHIP system. A total of 25 patients (26 hips) from June 2019-December 2023 met the selection criteria for inclusion in the study.

Population Study

There were 13 males and 12 females in this group; the ages ranged from 44 to 90 years, with an average of 69.1 years. There were 10 hips on the left side and 16 on the right side. There were 22 initial revision hips, 3 s revision hips, and 1 third revision hip. The reasons for revision were as follows: 14 hips with loose prosthesis, 4 hips with loose acetabular cup, 3 hips with osteolysis, 2 hips with acetabular dislocation, 1 hip with postoperative infection, 1 hip with prosthesis wear and tear, and 1 hip with periprosthetic fracture. Except for 1 patient who had spinal ankylosis with abnormal spinal curvature, the remaining patients had normal spinal curvature and no spinal ankylosis. The hip impingement test of the affected side was positive, and Trendelenburg's sign was positive in 16 hips and negative in 10 hips. Paprosky's acetabular bone defect was staged as follows: 7 hips in Stage IIA, 10 hips in Stage IIB, 5 hips in Stage IIC, 3 hips in Stage IIIA, and 1 hip in Stage IIIB. See Table 1.

Table 1 Clinical data of the patients

Preoperative planning using the AIHIP system

①The patient's pelvic orthopantomogram and 256-row CT scanning images of the hip joint were prepared before surgery, scanning of the entire pelvis and 15 cm below the small rotor of the femur was needed, and the thickness of the CT scanning layer was 0.8 mm. ② The scanning data were imported into AIHIP software in Dicom format, and then the three-dimensional reconstruction of the hip joint was completed through the Transformerunet algorithm. ③ Intelligent planning of the femoral and acetabular sides through the AIHIP system and placement of the acetabular cup prosthesis at an angle of 40° of acetabular abduction and 20° of anterior tilt, followed by selection of the appropriate ball head according to the reconstruction model to simulate of prosthesis placement. ④ Through the intelligent simulation of osteotomy, the retained length of the femoral spur and the distance from the tip of the greater trochanter to the shoulder of the femoral stem are measured, and finally the intelligent planning results are output to simulate the postoperative results. The preoperative planning time of the AIHIP system is approximately 5 min. see Figs. 1 and 2.

Fig. 1

Flowchart of the preoperative planning of the AIHIP system

Fig. 2

Individual examples of preoperative planning with the AIHIP system a AI calculation of acetabular cup placement; b-e Acetabular side details; f Simulated postoperative anterior 3D view; g Simulated postoperative posterior 3D view; h Simulated postoperative overall 3D view; i Simulated anteroposterior radiographs of the hip joint

Surgical methods

Surgeries are performed using conventional posterior lateral approach, resection of periarticular necrotic tissue and dislocation of the hip joint. According to the patient's specific situation, the corresponding prosthesis was removed, the residual bone cement and hyperplastic tissue were removed from the bone end, and the loosened part of the original prosthesis was replaced with a suitable type of planning prosthesis. During the operation, according to the acetabular bone defect, an appropriate amount of 5 mm diameter allograft pellet bone or metal pads and other fillers were implanted. In this group, the ABC plan (from mild to severe) was adopted for the revision surgery: ① A plan: metallic trabecular acetabulum/porous cup + pellet bone implant + metal pad if necessary; 24 hips in this group. ② Scheme B: Granulated bone graft + cemented polyethylene liner + metal reinforcement ring (cup-cage technique); 1 hip in this group. ③ Option C: 3D printed customized acetabular prosthesis + granular bone graft; 1 hip in this group. For the treatment of lateral femoral prostheses, 160 long shanks (3 hips) or solution long shanks (7 hips) from Johnson & Johnson (USA) were used for revision by Beijing CHUN Li Medical Instrument Co. In the process of loosening the femoral stem or implanting the femoral stem prosthesis, it is very easy to cause proximal femur fracture, and sternal wire binding and fixation were performed during the operation. In this group, the acetabular liner and femoral head were replaced alone in 5 hips, the acetabular cup, liner and artificial femoral head were replaced in 9 hips, the femoral stem and femoral head were replaced alone in 5 hips, and the acetabular and femoral stem prosthesis was completely revised in 7 hips.

Perioperative management

Patients with underlying diseases (e.g. diabetes mellitus, hypertension, etc.) can routinely stop their medications one day before surgery, and patients taking anticoagulants (e.g. reserpine) before surgery need to stop their medications two weeks before surgery. The patients' vital signs were closely monitored 24 h after the operation, and the recovery of consciousness, sensation and movement of both lower limbs were observed. And after surgery, patients were routinely given pantoprazole to protect the stomach, ambroxol to reduce phlegm, citrus flavonoid tablets to reduce swelling, flurbiprofenate injection for analgesia, ceftriaxone sodium to fight infection, enoxaparin sodium for anticoagulant, etc., and dexuzumab injection, etc., as appropriate, was given to osteoporotic patients. One- to three-day postoperative patients were asked to perform medial femoral muscle contraction and foot dorsiflexion and plantar flexion activities, consisting of 5 s of restraint and then relaxation, 20–30 times/group, 3–4 groups/d. In simple acetabular side revision patients, only the replacement of the artificial femoral head or the acetabular cup liner, patients were asked to perform leg lifting training at the bedside, and according to the patients' conditions, they gradually carried out training for getting off the bed and walking, until the patients could get out of the walkers to walk under load; the patients who had total revision of the femoral and acetabular side prostheses were asked to perform leg raising training at the bedside, and according to the patients' situation, the patients could walk without walking aids. For those with total revision of the femoral and acetabular prostheses, the bedside exercise is the main focus, followed by gradual bedrest and walking after 3 months of bed rest.

Observational indicators of efficacy

The patients' replacement prosthesis models, operation time, hospitalization time, and postoperative time out of bed were recorded, as well as the occurrence of adverse events such as postoperative infection, fracture, and loosening of the prosthesis. The HHS was used to evaluate the function of the affected limbs before surgery, and 1 week and 6 months after surgery; hip mobility was compared preoperatively and 6 months after surgery. Among them, for prosthesis matching: the postoperative prosthesis model was exactly the same as the preoperative design as a perfect match, the difference of prosthesis model between pre- and post-surgery by 1 number as an average match, and the difference in prosthesis model between pre- and post-surgery by 2 or more numbers as a mismatch [11]. Prosthesis position: Postoperative review of the orthopantomogram of the hip joint and measurement of the prosthesis position, revealed that the femoral prosthesis was in the center of the fixation position between 3° and 3° of internal rotation, the acetabular prosthesis abduction angle was in the range of 30° to 50°, the anterior tilt angle was between 5° and 25°, and the acetabular prosthesis could be considered to be in the safe range [13, 14].

Statistical methods

SPSS 25.0 statistical software was used for analysis. Normally distributed data were analyzed by the Shapiro‒Wilk test, and the data are expressed as the mean ± standard deviation. The comparison of hip mobility before and after surgery was performed by paired t tests. The comparison of Harris scores was performed by one-way repeated measures ANOVA, and if the sphericity test was not satisfactory, it was corrected by the Greenhouse–Geisser method, and comparisons between different time points were performed by the Bonferroni method; test standard α = 0.05.

Results

In this group, the operation time ranged from 85 to 510 min, with an average of 241.8 min; the hospitalization time ranged from 7 to 35 days, with an average of 15.2 days; and the time of discharge from the walker ranged from 2 to 108 days, with an average of 44.7 days. Twenty-five patients were followed up for 6 to 59 months, with an average of 25.3 months. Except for one patient who had a thigh hematoma (treated with incision and drainage and decompression) and one with hip dislocation (repaired), none of the other patients experienced any adverse events, such as loosening of the prosthesis or infection. The postoperative acetabular cup type matching degree completely matched 25 hips, not matching 1 hip (+ 2), for a matching rate of 96.15%; the femoral stem type matching degree completely matched 25 hips, generally matching 1 hip (−1), for a matching rate of 100%. The Harris scores at 1 week and 6 months after surgery were 54.7 ± 9.6 and 89.6 ± 7.0, respectively, which were significantly improved compared with the preoperative scores of 33.5 ± 8.3, and the difference was statistically significant at 6 months after surgery compared with the 1-week period (P < 0.05). The patient's hip function was evaluated according to the Harris scores at 6 months after surgery, and the patients were awarded good 23 hips and medium 3 hips, which can meet the needs of daily life. Hip mobility at 6 months post-surgery was 111.15 ± 9.72°, and the difference was statistically significant (t = −8.077, P < 0.001) compared with the preoperative value of 79.42 ± 17.51° (Figs. 3, 4, 5, 6 and 7 and Table 2).

Fig. 3

Matching of AI-designed prostheses with actual surgical prostheses

Fig. 4

Harris score and changes in joint mobility

Fig. 5

Patient, 74 years old, postoperative loosening of the bioprosthesis with periprosthetic fracture of the lateral femur, major revision of the lateral femur a. Preoperative; b. Installation of cemented mortise cups; c. Splitting of the lateral femoral muscle to expose the distal femur; d. Kirschner's pin drilling to open the window; e. Postoperative

Fig. 6

Patient, male, 46 years old, right hip Paprosky stage IIC bone defect with total revision of hip prosthesis a. Preoperative; b. Preoperative CT sagittal; c. Postoperative

Fig. 7

Patient, female, 75 years old, biologic prosthesis, primary revision acetabular side a. Preoperative; b. Preoperative CT coronal; c. Postoperative

Table 2 Comparison of postoperative Harris scores with preoperative scores

Discussion

The AIHIP system has shown great advantages in THA preoperative planning, and an increasing number of hospitals in China have adopted AI-assisted THA preoperative planning. The results of this study show that AI preoperative planning has a high accuracy in predicting prosthesis models in hip revision surgery. By analyzing the efficacy of postoperative hip revision surgery assisted by the AIHIP system in this study, we summarize the advantages of this system as follows:

First, the AIHIP system can improve the accuracy of preoperative planning. The AIHIP system adopts a unique Transformerunet algorithm, which can automatically and accurately segment CT images of revision hip joints in a short period of time to improve clinical efficiency, and has high practicality and clinical application value [15, 16]. Compared with traditional two-dimensional and three-dimensional planning, preoperative planning using the AIHIP system results in more accurate magnification of preoperative X-ray film measurements, less variability in measurement angles, more accurate prosthesis models, and easier operation, thus improving surgical efficacy [17,18,19]. Second, the AIHIP system reduces the intraoperative error rate. Using the AIHIP system to design a lateral femoral prosthesis model in three dimensions can largely circumvent the problems of manual measurement and time consumption in the traditional preoperative two-dimensional design, improve clinical efficacy, and reduce the postoperative revision rate. Most of the acetabular lateral bone defects in the hip revision patients included in this study were Paprosky stage II and III defects, which often increase the risk of surgical failure due to the small amount of residual bone, poor bone quality, and small contact area between the bone and the acetabular cup, resulting in the lack of a match between the acetabular cup prosthesis support points. If the intraoperative acetabular polishing is shallow, the acetabular cup is unstable, and the prosthesis can easily to loosen after surgery; if the intraoperative acetabular polishing is deep, it is easy to penetrate the posterior wall of the acetabulum, which directly leads to the failure of hip joint revision surgery [20]. In the face of complex revision surgery, preoperative planning using the AIHIP system can simulate the range of motion, thus optimizing the placement of the acetabular cup, preventing impingement, acetabular cup prosthesis displacement, postoperative massive bone defects and screw failure, and reducing the occurrence of postoperative complications [21, 22]. Third, the AIHIP system can improve surgical efficiency. To use the AIHIP system for preoperative planning, it is only necessary to prepare the corresponding imaging data and scan them into CT data beforehand, and then import these data in Dicom format; the AIHIP system can automatically complete 3D reconstruction of the hip joint. This study revealed that the preoperative planning time of the AIHIP system is approximately 5 min, which is accurate and rapid, and improves its clinical efficacy.

Currently, AI preoperative planning and design is increasingly common in other joint fields in addition to its application in hip revision, and the integration of artificial intelligence (AI) into decision support systems for the diagnosis and treatment of orthopaedic diseases is one of the main directions in the development of orthopaedic technology [23]. Long Wu et al. [24] showed that in the AI planning group, the complete accuracy rate of acetabular cup prostheses was 54%, and the complete accuracy rate of femoral stem prostheses was 64%, indicating that AI preoperative 3D planning predicts prosthesis models significantly more accurately than the traditional X-ray film template method. In knee surgery, AI preoperative planning is also widely used, and studies have shown [25] that in total knee arthroplasty the accuracy of prosthesis size prediction in the AI group was significantly higher than that in the 2D group, with the complete compliance rates of femoral and tibial prostheses in the AI reconstruction group of 90% (27/30) and 86.7% (26/30), respectively. the corresponding rates in the 2D template group were 66.7% (20/30) and 60% (18/30), with high prediction accuracy. Another study showed [26] that AIJOINT had an accuracy of 92.9% in planning femoral and tibial component sizing in total knee arthroplasty, which was significantly higher than the accuracy of 42.9% and 47.6% in the conventional approach. In addition, AI preoperative planning also improved the accuracy of hip-knee-ankle angulation and reduced postoperative blood loss. Germann et al. [27] showed that using automatic identification of anterior cruciate ligament (ACL) tears had high accuracy with 99% sensitivity, 94% specificity, and 97% AUC, using arthroscopic surgery as the reference standard. Moreover, preoperative planning using AI technology can improve the accuracy of tunnel placement during ACL surgery and help to assess kinematics and stability after ACL reconstruction (ACLR) [28].

However, the AIHIP system still has limitations. First, the AI preoperative planning technology is not mature enough to perform well for severe acetabular side defects, and the AIHIP system is currently unable to design options such as multiple cushions, reinforcing rings, customized prostheses, or customized reinforcing rings, whereas patients who require revision surgery are usually more complex and require more revision options. Second, the AI preoperative planning techniques are not yet fully clinically compatible. For example, when AI designs a porous cushion, the size of the cushion may not be perfectly matched with the actual surgery, which may affect the smoothness of the operation; and even if the cushion installation position and angle are simulated before the surgery, several trial molds are still needed during the surgery to achieve the maximum degree of adaptability. Third, it is important to note that soft tissue conditions, specific surgical access routes and intraoperative prosthesis management are major difficulties in AI assessment. The diversity and unpredictability of these factors make it a major challenge for AI systems to provide reliable assessment and decision support, further limiting their potential for application in real-world clinical settings. Huo et al. [22] reported that the AI preoperative planning system may produce large errors in the face of high hip dysplasia. For the management of bone defects, the selection of an appropriate prosthesis is the key to successful hip revision surgery [29]. Paprosky stage IIA and IIB defects are mostly due to bone loss in the anterior and superior acetabulum, and the acetabular cup is displaced anteriorly superiorly or posterolaterally, which usually does not require additional filling of allografts [30]. Stage III defects, in addition to the need for filling of allografts, need to be filled with CT three-dimensional reconstruction and 3D printing In addition to the need for allograft filling, stage III defects also require the use of CT 3D reconstruction and 3D printing to design a prosthesis implantation plan that best matches the pelvis and surrounding organs, and we mostly adopt the cup-cage technique or 3D printing to design personalized prosthesis implantation [31,32,33,34], to promote the stabilization of the acetabular prosthesis. The compression implant technique can be simple and effective in the management of acetabular nonstructural bone defects [35], but for severe structural bone defects, customized prostheses are needed. In addition, the AIHIP system, like any other AI, is still unable to avoid the appearance of errors (e.g., AI hallucinations, etc.) in the results it produces, and the specifics still need to be judged by the surgeon based on the actual intraoperative situation.

In summary, the AIHIP system is used to assist in the preoperative design of revision surgery for THA patients with a high degree of accuracy, which can reduce the difficulty of surgery and promote patient recovery. In addition, the use of AIHIP in preoperative planning enables a highly personalized and dynamic approach, and its continuous learning capabilities help to continuously improve predictions and recommendations and optimize long-term outcomes. The benefits of adopting AI technology should not be limited to preoperative planning, but should also permeate the postoperative phase. Patient satisfaction is considered as the primary outcome, and the patient's postoperative rehabilitation program is intelligently adapted [36]. With the increasing interest of orthopedic surgeons, research on AI tools applicable to the management of revision hip surgery is growing exponentially and is expected to have a significant future impact on its clinical application [37].

However, there are still limitations in this study: (1) due to the limitations of Paprosky stage II-III bone defects in adults, the number of patients who needed revision surgery was relatively small, so the number of patients in this study was relatively small; (2) the follow-up time of patients was limited, and the long-term effects need to be clarified by further follow-up; and (3) there were large individual differences among revision patients, and the revision protocols were not the same.

Conclusion

We believe that relying on the AIHIP system for preoperative planning can help operators perform such complex and difficult hip surgeries more invasively, intelligently, safely, and accurately, and it is worthwhile to further explore how to further improve the ability of the AIHIP system to address aspects such as soft tissue segmentation and hip dysplasia design.