BMC musculoskeletal disorders · 2026
Optimizing preoperative planning for total hip arthroplasty using random forest models to predict stem size and compatibility.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Kaneoka Takehiro, Imagama Takashi, Okazaki Tomoya, Matsuki Yuta, Kawakami Takehiro, Yamazaki Kazuhiro, Sasaki Kei, Asai Yoshiyuki, Sakai Takashi
Аннотация
BACKGROUND: Preoperative planning is essential for optimal outcomes post-total hip arthroplasty (THA). Appropriate stem size and compatibility with femoral dimensions are essential in THA to prevent distal fixation. Previous studies have proposed deep learning-based models with image recognition for preoperative planning to determine the best-fit stem; however, whether size and compatibility are adequately optimized remains unclear. This study developed methods that may assist surgeons in preoperative planning to estimate the stem size and compatibility using supervised machine learning models.
METHODS: Two Random Forest (RF) models were developed to estimate the best-fit stem size and stem compatibility with the femoral geometry. Femoral size information measured at 10 locations from computed tomography images of 320 hips was used to train the models. The training data included the sizes and compatibility information as the target variables for supervised learning provided by a simulation software currently used in clinical practice to determine the optimal stem. As part of feature engineering, ratios derived from combinations of the 10 measured values were used as learning data. The size estimation model was designed as a seven-class classification model, whereas the compatibility estimation model was a binary classification model, predicting whether the stem will exclusively fit distally or not. Both models were tested using data from 109 hips of patients who underwent THA. The model performances were assessed using accuracy, F1 score, precision, and recall. Each parameter's impact was determined using feature importance analysis.
RESULTS: The size and compatibility estimation models trained without ratio information showed accuracies of 89.0% (exact-match: 46.8%) and 85.3%, respectively. When it was included, the models' performance improved, with the size and compatibility estimation accuracies increasing to 92.7% (exact-match: 39.4%) and 87.2%, respectively. Feature importance analysis highlighted that distal medullary cavity diameter is key in size estimation, while overall femoral dimensions are vital for compatibility estimation.
CONCLUSIONS: This study developed models that estimate the two critical factors in stem selection: size and compatibility. These models may support preoperative planning for THA using the Accolade II stem, although further validation is required to establish applicability to other implant systems.
METHODS: Two Random Forest (RF) models were developed to estimate the best-fit stem size and stem compatibility with the femoral geometry. Femoral size information measured at 10 locations from computed tomography images of 320 hips was used to train the models. The training data included the sizes and compatibility information as the target variables for supervised learning provided by a simulation software currently used in clinical practice to determine the optimal stem. As part of feature engineering, ratios derived from combinations of the 10 measured values were used as learning data. The size estimation model was designed as a seven-class classification model, whereas the compatibility estimation model was a binary classification model, predicting whether the stem will exclusively fit distally or not. Both models were tested using data from 109 hips of patients who underwent THA. The model performances were assessed using accuracy, F1 score, precision, and recall. Each parameter's impact was determined using feature importance analysis.
RESULTS: The size and compatibility estimation models trained without ratio information showed accuracies of 89.0% (exact-match: 46.8%) and 85.3%, respectively. When it was included, the models' performance improved, with the size and compatibility estimation accuracies increasing to 92.7% (exact-match: 39.4%) and 87.2%, respectively. Feature importance analysis highlighted that distal medullary cavity diameter is key in size estimation, while overall femoral dimensions are vital for compatibility estimation.
CONCLUSIONS: This study developed models that estimate the two critical factors in stem selection: size and compatibility. These models may support preoperative planning for THA using the Accolade II stem, although further validation is required to establish applicability to other implant systems.