3种人工神经网络模型预测大肠癌的初步研究

Application of three artificial neural network models in diagnosing colorectal carcinoma

目的探讨大肠癌血清蛋白标志物、肿瘤标志物及联合多标志物人工神经网络(ANN)模型在预测大肠癌中的价值。方法大肠癌与健康对照血清样本106例,利用表面增强激光解吸电离飞行时间质谱(SELDI-TOF-MS)检测血清蛋白质谱并筛选大肠癌蛋白标志物,电化学发光法检测癌胚抗原(CEA)、甲胎蛋白(AFP)、糖类抗原72-4(CA72-4)和CA19-9分别建立蛋白标志物、肿瘤标志物及蛋白标志物与肿瘤标志物结合的多标志物ANN模型。结果大肠癌患者和对照组间比较差异有统计学意义(P<0.001),筛选4个质荷比(m/z)分别为4 095、5 6404、4807、620 m/z蛋白建立ANN模型,预测大肠癌敏感度和特异度为92.3%和83.3%;肿瘤标志物模型预测大肠癌的敏感度为73.1%,特异度86.7%;联合筛选的4个标志蛋白和CEA、CA72-4建立的模型诊断大肠癌敏感度和特异度分别为92.3%和96.7%。结论联合蛋白标志物和肿瘤标志物建立ANN模型,在预测大肠癌中显示高通量、高敏感性和高特异性的特点,具有潜在应用价值。

Objective To establish three artificial neural network（ANN）models based on protein markers,tumor markers,multi-markers of colorectal carcinoma and discuss the application of these in diagnosing colorectal carcinoma.Methods This study included 106 serum from patients with colorectal carcinoma and healthy controls.The SELDI-TOF-MS were performed to screen serum protein markers.The ANN model based on the screened protein markers was established.The levels of carcinoembryonic antigen（CEA）,AFP（α-fetoprotein）,CA19-9（cancer antigen）and CA72-4 in serum from patients with colorectal carcinoma were detected by electrochemiluminescence determination.The model based on the tumor markers was established.The multi-markers model was constructed by the screened protein markers and tumor markers.Results The result showed a significant difference between patients with colorectal carcinoma and healthy controls（P0.01）.Four specific protein peaks（m/z at 4 095,5 640,4 480,7 620） were chosen to build an ANN model,the model showed a sensitivity of 92.3%,a specificity of 83.3% to diagnose the colorectal carcinoma.The ANN model of tumor markers had a sensitivity of 73.1% and a specificity of 86.7% to diagnose the colorectal carcinoma.The multi-markers model showed a sensitivity of 92.3%,a specificity of 96.7% to diagnose colorectal carcinoma.Conclusion The preliminary results suggest that the multi-markers model based on protein markers and tumor markers have the high sensibility and specificity in diagnosing colorectal carcinoma.