血液可见吸收光谱与血氧参数神经网络估算法

总血红蛋白浓度和血氧饱和度是两个基本的血氧参数。文章提出了利用内置双光纤微创探头在位测量大鼠脑组织血氧参数的新方法。首先,利用悬乳液(Intralipid)和全血配置不同总血红蛋白浓度的混合溶液,模拟生物组织模型,用光纤光谱仪测试系统测量组织模型在加氧和去氧时的实时吸收光谱,同时用血氧分析仪(OXI meter)对血氧参数定标,建立测试光谱和定标数据样本集。然后,利用人工神经网络建立血液吸收光谱与血氧参数的神经网络模型,训练后的网络模型能根据吸收光谱输出生物组织的血氧参数值,总血红蛋白浓度和血氧饱和度的平均输出误差分别为±4μmol·L-1和±5%。最后,利用神经网络模型对大鼠脑组织血氧参数进行了在位测试实验,测得脑灰质的血氧饱和度为0.60~0.70,脑白质血氧饱和度为0.45~0.55;总血红蛋白浓度在脑皮层(深度1mm)附近最高,平均110μmol·L-1,其余深度脑组织的总血红蛋白浓度为70~90μmol·L-1。这种方法对脑外科微创手术中实时在位测试脑组织血氧参数具有重要的参考意义。

Estimation of Tissue's Blood Oxygen Parameters from Visible Absorption Spectrum of Tissues by Artificial Neural Network

Total hemoglobin concentration (THC) and hemoglobin oxygen saturation (SO2) are essential parameters to doctors who wonder patients' hematogenous conditions and oxygen supplies and consumptions. Instruments presently used for measuring these parameters have big size of detecting probes that limit their applications to inner bodies. An optical probe involving two fibers with source-detector separations of one hundred micrometers was developed in the present study for purpose of minimally invasive inner detecting, which uses steady-state, broadband (300-1 000 nm) light source. The source light is delivered to targets through one fiber and the reflected light from the targets is collected and transferred to a spectrometer through the other fiber. Reflectance spectrum is obtained from the spectrometer. The method of reading THC and SO2 from the reflectance spectrum was developed using liquid-tissue phantoms containing intralipid and blood. Firstly, reflex spectrum of intralipid was recorded before mixtures of intralipid and blood with different THC were made as tissue phantoms. Then the fiber optical spectrometer was used to obtain reflex spectra as the phantoms' SO2 changed; simultaneously their corresponding THC and SO2 were recorded as the scale values by an oximeter. Differences of reflex spectra in 520-590 nm between intralipid and tissue models were proved reliably. Secondly, after data collections of absorption spectra and scale values were finished, two artificial neural networks (ANN) were build to model the relationship between scale values and absorption spectra. After being trained, the ANNs could output THC and SO2 correctly when an absorption spectrum was input. The ANNs produced errors of less than 4 micromol x L(-1) for THC and 5% for SO2. In vivo and minimally invasive measurements of THC and SO2 of brain tissues in different depth were finished on 30 rats by this specific system with the ANNs. The probe was inserted stereotactically to a depth of 6 mm with measurements obtained every 0.2 mm. SO2 of gray mater and white mater of rats was respectively obtained as 0.60-0.70 and 0.45-0.55. The highest THC, 110 micromol x L(-1) was measured around rat cortex. THC of brain tissue in other depth is 70-90 micromol x L(-1). These values agree well with reported data. This simple, inexpensive method deserves further study to establish its efficacy for THC and SO2 measurements of inner body.