Publication:

生物阻抗與類神經網路應用於優秀運動員的身體組成分析

生物阻抗分析法（BIA）可簡便、快速、無侵入性地估測身體組成。研究指出估測方程式是造成BIA不準確的主要原因，更建議要依受測者條件選擇適合的估測方程式。

本篇研究以24名台灣體院男子足球隊員為受測者（年齡20.3±1.9歲，身高173.6±5.6 cm，體重66.1±5.3 kg），以雙能X光吸收儀（DEXA）測量之身體組成為效標，比較生物阻抗分析儀（BC-418）所估算出的去脂肪質量（FFM），以BIA所測得的阻抗值搭配年齡、身高及體重等參數分別以線性迴歸（LR）的迴歸方程式及類神經網路（ANN）估測身體組成。結果顯示，LR的r2=.897，RMSE=1.678 kg；五個神經元ANN的r2=.996，RMSE=0.328 kg，ANN優於LR。LR與ANN對DEXA的FFM偏差值（bias）均約為0 kg，BC-418對DEXA的bias= - 0.628 kg，LR與ANN優於BC-418的內建估測方程；LR、BC-418及ANN對DEXA的FFM偏差範圍（2 S.D.）分別為3.357、3.958及0.656 kg，ANN優於LR及BC-418。結論：本篇研究以類神經網路估測足球員身體組成優於線性迴歸及生物阻抗分析儀（BC-418）內建方程的估測結果。

Bioelectrical impedance analysis (BIA) can estimate body composition easily, rapidly and non-invasively. Some papers have indicated that the accuracy of predictive equations of BIA mainly depend on the equations itself, even, the specific subjects need specified equation.

The purpose of this study was to estimate body composition of the football players with the BIA measurement compared to Dual-energy X ray absorptiometry (DEXA).Method: Subjects, 24 football players of National Taiwan College of Physical Education, with mean age at 20.3±1.9 years, mean height at 173.6±5.6 cm and mean weight at 66.1±5.3 kg. To evaluate the accuracy of predictive fat-free mass (FFM) of body composition by bioelectrical impedance analyser (BC-418), DEXA, as criteria method, was compared. By the measured data as factors including the bioelectrical impedance values (Z) of hand-to-foot modulation by BIA in right side, gender, age, height, and weight, the predictive equation by traditional linear regression analysis (LR) for FFM by DEXA was gained, also, the ANN predictive model created.

Result: The lower r2 and greater RMSE in LR (r2=.897 and RMSE=1.678 kg) than in ANN (r2=.996 and RMSE=0.328 kg) were gained. ANN is better than LR. The biases that FFM of LR and ANN compare to DEXA are about 0 kg. The bias that FFM of BC-418 compares to DEXA is -0.628 kg. LR and ANN are better than BC-418. The FFM’s range of bias (2 SD.) of FFM of LR, BC-418 and ANN compared to DEXA are 3.357, 3.958 and 0.656 kg, respectively. ANN is better than LR and BC-418.

Conclusion: To estimate body composition of football players, ANN is more applicable than LR and estimated equation of BC-418.