To investigate the differences between a deep learning model and manual measurement in assessing postoperative parameters of intertrochanteric femoral fractures after intramedullary nail fixation.

A retrospective study was conducted, including 150 patients with intertrochanteric femoral fractures who met the inclusion criteria and underwent unilateral intramedullary nail fixation at the Department of Orthopaedics and Traumatology, Beijing Jishuitan Hospital, Capital Medical University, from January to December 2024. Among them, 50 patients each in the proximal femoral nail antirotation (PFNA), Gamma-type intramedullary nail, and InterTan intramedullary nails groups. There were 81 males and 69 females, aged (64.1±17.3) years; 79 cases were on the left side and 71 on the right. The deep learning model and manual method were used to measure the postoperative neck-shaft angle, tip-apex distance (TAD), and medial femoral cortex support, and the results were compared. Bland-Altman analysis was used to evaluate the agreement between the two measurement methods.

For TAD, there was no statistically significant difference between the deep learning model and manual measurement in all cases and each subgroup (P>0.05). The overall mean difference was (-0.21±5.29) mm, and Bland-Altman analysis showed good agreement. The overall Kappa value for the measurement of medial femoral cortical support was 0.716, with the best agreement in the PFNA group (Kappa=0.803), followed by the Gamma-type intramedullary nail group and InterTan intramedullary nail group (0.675 and 0.651, respectively). The overall neck-shaft angle measured by the deep learning model was higher than that measured by the manual method, especially in the InterTan intramedullary nail group, where the difference was statistically significant (P<0.001), while no statistically significant difference was found in the PFNA group or the Gamma-type intramedullary nail group (P>0.05).

The deep learning model showed good agreement with the manual method in measuring TAD and medial femoral cortical support, as well as postoperative neck-shaft angle in the PFNA and Gamma-type intramedullary nail fixation. However, there was a systematic bias in the measurement of postoperative neck-shaft angle in the InterTan intramedullary nail fixation, suggesting that the model still needs further optimization for complex dual-screw configurations.