机构地区: 韩山师范学院化学系
出 处: 《化学研究与应用》 2007年第4期420-423,共4页
摘 要: The probabilistic neural network(PNN) was applied in the recognition of cancerous stomach tissues.The characteristic FTIR peak frequencies(including vas(CH3),vs(CH2),δ(CH2),vas(PO2-),v(C-O),vs(PO2-) and vs(nucleic acid(DNA,RNA),cell proteins and membrance lipids) from corresponding stomach cancer tissues were used as the input vectors of the PNN neural network.The experimental results as follows: 1) when net parameter,spread,is 0.25~2.55,and the vas(PO2-),v(C-O),vs(PO2-) and vs(nucleic acid(DNA,RNA),cell proteins and membrance lipids) were employed as input vectors,PNN neural network exhibited the best performance with a mean accurate rate of recognition up to 96.2%;and when the spread between 0.2~4.60,and all the characteristic FTIR peak frequencies mentioned above were used as net input vectors,the PNN neural network exhibited a slightly lower performance with a mean accurate rate of recognition of 92.3%,while the PNN neural network performed very poorly when just v(C-O) and vs(PO2-) were used as input vectors with a mean accurate rate of recognition of 53.8%.2) In general,in comparison,PNN neural network should obviously be more excellent with respect to LVQ neural network(see ref.) in prediction and recognition of cancerous stomach tissues. The probabilistic neural network (PNN) was applied in the recognition of cancerous stomach tissues. The characteristic FTIR peak frequencies ( including vas ( CH3 ), vs ( CH2 ), δ ( CH2 ), vas ( PO2ˉ ), v ( C - O), vs ( PO2ˉ ) and vs ( nucleic acid (DNA, RNA), cell proteins and membrance lipids) from corresponding stomach cancer tissues were used as the input vectors of the PNN neural network. The experimental results as follows: 1 ) when net parameter, spread, is 0.25 - 2.55, and the vas ( PO2ˉ ) ,v( C - O) ,vs ( PO2ˉ ) and vs ( nucleic acid ( DNA, RNA), cell proteins and membrance lipids) were employed as input vectors, PNN neural network exhibited the best performance with a mean accurate rate of recognition up to 96.2% ; and when the spread between 0.2 - 4.60, and all the characteristic FTIR peak frequencies mentioned above were used as net input vectors, the PNN neural network exhibited a slightly lower performance with a mean accurate rate of recognition of 92.3%, while the PNN neural network performed very poorly when just v( C - O) and v' (PO2 - ) were used as input vectors with a mean accurate rate of recognition of 53.8%. 2) In general, in comparison, PNN neural network should obviously be more excellent with respect to LVQ neural network (see ref. [ 10] ) in prediction and recognition of cancerous stomach tissues.