Performance of artificial intelligence in automated measurement of patellofemoral joint parameters: a systematic review

Comprehensive Summary

This study conducts a systematic review to evaluate how artificial intelligence performs in automatically measuring patellofemoral joint parameters for patellar dislocation diagnosis. 19 studies with 10490 patients from databases including PubMed, Web of Science, Cochrane Library, and Embase were included in the study. Six studies developed AI models using radiographic series, nine used CT imaging and four used MRI. The AI models showed generally good agreement with manual measurement done by experts, with greater efficiency. Intraclass correlation coefficients (ICCs) ranged from 0.9 to 0.94 for femoral anteversion angle, 0.91 to 0.92 for trochlear groove depth and 0.93 to 0.95 for tibial tuberosity-trochlear groove distance, and 0.86 to 0.91 for trochlear dysplasia. Other measurements still showed great reliability such as patellar height with ICCs of 0.88 to 0.985, sulcus angle with ICCs of 0.878 to 0.98, patellar tilt angle with ICCs of 0.79 to 0.99. These measurements all showed great accuracy that holds up to measurements done by experts with great efficiency, indicating possible integration into clinical support systems.

Outcomes and Implications

Patellofemoral joint is a point of interest in orthopedics as patellar dislocation is one of the most difficult conditions to treat. The patient experiences frequent dislocation, knee pain and instability, impacting their quality of life drastically. Moreover, accurate assessment of patellofemoral joint parameters is required for accurate diagnosis which is time-consuming, inconsistent and demands expertise by the radiologist or orthopedist. Therefore, application of AI in this field would be helpful in saving much needed resources and potentially enhancing the consistency of diagnosis results. Many studies have been conducted on the matter and a systematic review would help evaluate AI’s performance in this field.