Sex estimation from anteroposterior knee radiographs in a Brazilian sample

Abstract Background Radiographic measurements of the bones forming the knee may serve as an alternative method for sex estimation in forensic contexts, particularly in the examination of human remains. This study aimed to assess whether linear measurements derived from anteroposterior knee radiographs could be used for sex estimation in a Brazilian sample using binary logistic regression (BLR) models. The sample comprised 474 left knee radiographic images (244 females and 230 males) from individuals aged 18 years or older. For the analysis, two measurements were taken at the distal femoral epiphysis and one at the tibia. Univariate BLR models were constructed for each measurement. Internal validation used 10-fold cross-validation to assess model stability and generalizability. Results In the final analytical sample, the univariate BLR model based on the TPML measurement showed the best predictive performance, achieving an accuracy of 81% during 10-fold cross-validation, followed by the FBEB (78%) and FCML (63%) models. Complementary analyses using LDA and decision tree models achieved accuracies of 77% and 82%, respectively. Conclusion Anteroposterior knee radiographs demonstrated potential for sex estimation in a Brazilian sample, with the TPML measurement showing the best predictive performance. The proposed models may serve as an auxiliary tool in forensic investigations involving radiographic records. Nevertheless, further studies using calibrated images and external validation in independent populations are needed to strengthen the robustness, generalizability, and forensic applicability of these models.