Direct to digital oximeter and method for calculating oxygenation levels

[0002] This application is a division of application Ser. No 08/201,942, filed Dec. 1, 1998, now U.S. Pat. No. 6,163,715, which is a continuation-in-part of application Ser. No. 08/683,617, filed Jul. 17, 1996, now U.S. Pat. No. 5,842,981.

[0003] The present invention is concerned generally with an improved oximeter for non-invasively measuring arterial oxygen saturation. More particularly, this invention is concerned with an improved method for direct digital signal formation from input signals produced by a sensor device which is connected to the oximeter.

[0004] In all oximeters, input signals are received from a sensor device which is directly connected to the blood-carrying tissue of a patient, such as a finger or ear lobe. The sensor device generally consists of a red LED, an infrared LED, and one or two photodetectors. Light from each LED is transmitted through the tissue, and the photodetectors detect the amount of light which passes through the tissue. The detected light consists of two components for each bandwidth. An AC component represents the amount of pulsating blood detected, while the DC component represents the amount of non-pulsating blood. Therefore, four separate components of detected light are examined in order to determine the arterial oxygen saturation: red DC, red AC, infrared DC and infrared AC. The amount of light detected is then used to determine the oxygen saturation in the blood of the patient based on the following equation:

[0005] (IR(AC)/IR(DC))/(Red(AC)/Red(DC))

[0006] In a traditional oximeter, the sensor output signal is converted to an analog voltage and then separated into infrared and red components. Some oximeters further separate the AC and DC components. Separate analog circuits are then used to sample, demultiplex, and filter these signals. In these systems, therefore, it is necessary to carefully match the analog components to minimize errors which can result from differences in gain or frequency response in the two circuits. Furthermore, because of the need to carefully match hardware for each analog input circuit, and the increased probability of errors when additional analog channels are added, traditional oximeters are generally limited to two analog inputs.

[0007] Additionally, the analog circuitry employed in traditional oximeters is generally insufficient to accurately detect low level signals. Therefore, these oximeters are generally ineffective for monitoring fetal conditions, as well as for use with patients with thick or very dark skin. Furthermore, the methods used in prior art oximeters for measuring oxygenation levels rely heavily on pulse detection and peak-valley measurements which are highly susceptible to variations due to motion artifact noise.

[0008] The instant invention improves on the analog signal processing employed in prior art oximeters by receiving input current signals from at least two and preferably three light emitting devices of different wavelengths and converting these input signals directly to digital voltage values, without first converting to analog voltages or separating the signals. This is accomplished by using a charge digitizing analog to digital converter with sufficient range to represent the large DC signals and sufficient resolution to represent the small AC signals. This charge digitizing converter employs a current integrator as the front stage, which tends to average and filter input noise. This is an improvement over the analog current to voltage conversion used in traditional oximeters, which tend to amplify noise.

[0009] Once the input current is converted to a digital voltage value, all input signals are processed along the same digital hardware path, instead of the separate analog hardware paths required by the traditional method. This system eliminates the need to match analog hardware components, and therefore further reduces potential errors. Furthermore, once the signals are digitized, a microprocessor can perform all of the signal processing, demultiplexing, and filtering steps required by traditional oximeters. This reduction in the analog signal processing stage increases both the speed and accuracy of the oximeter, decreases cost by eliminating expensive analog components, and reduces the size of the oximeter by eliminating physically large analog components.

[0010] In another aspect of the invention, a method for analyzing oxygenation levels without the need for pulse detection and peak-valley measurements is also disclosed. The method comprises the steps of storing vectors of contiguous, paired infrared and red data samples over a period of time, using a least-squares minimization method for determining an infrared to red ratio, and determining a noise metric for filtering noise from the resultant oxygenation calculations. The noise metric substantially filters noise due to motion artifact, such as source: detector geometry variations and respiration noise, thereby providing a more accurate oxygenation level reading.

[0011] In a further improvement, additional wavelengths can be added to the oximeter to improve noise filtering or add medical monitoring functions to the oximeter. Because all signal conversion is time-multiplexed through a single analog to digital converter circuit, a third or further wavelengths can be easily and inexpensively added to the sensor and device. The additional wavelengths can be used in a number of applications which increase the accuracy of the oximeter or provide additional monitoring functions, including: noise detection; dyshemoglobin detection and/or measurement; and indicator dye measurement.

[0012] Due to the ability of the digital circuitry of the present invention to process low level current input signals and to filter noise components and the additional noise filtering functions disclosed, the oximeter can be used to accurately monitor oxygenation levels which were previously difficult to monitor, including fetal oxygenation levels and the oxygenation levels of dark and thick skinned patients. In one particular embodiment the dynamic range of the analog to digital converter may be optimized to match the input signal range.

[0013] It is therefore an object of this invention to provide an improved method for non-invasively measuring fluid parameters.

[0014] It is another object of this invention to provide an improved method for measuring arterial blood saturation.

[0015] It is another object of the invention to provide improved speed and accuracy in the measurements provided by oximeters.

[0016] It is another object of the invention to provide a direct analog to digital conversion of the input current signal with sufficient range to measure large DC signals and enough resolution to represent small AC signals so that accurate measurements can be made with reduced analog signal processing.

[0017] It is another object of the invention to provide a reduction in potential errors by directly converting the input current signal to a digital voltage signal, thereby bypassing the current to voltage conversion step which can amplify noise.

[0018] It is another object of the invention to provide a reduction in potential errors by processing all signals along one digital hardware path, thereby eliminating the need for matched analog components.

[0019] It is another object of the invention to provide an improved oximeter having a reduced number of electronic circuit components.

[0020] It is still another object of the invention to provide a reduction in the size of oximeters by eliminating physically large analog components.

[0021] It is yet a further object of the invention to provide an improved method and system for directly converting to digital signal form at least two signals from light emitting devices of different wavelengths.

[0022] It is another object of the invention to provide an improved method for filtering noise from oxygenation level calculations.

[0023] It is yet another object of the invention to provide a dynamic range control for calculating oxygenation levels in a plurality of signal range levels.

[0024] It is still another object of the invention to provide an improved oximeter capable of monitoring a wide range of patients.

[0025] It is another object of the invention to provide a reduction in the size and cost of detecting more than two wavelengths in oximeters.

[0026] These and other object and advantages of the invention, together with the organization and manner of operation thereof, will become apparent from the following detailed description when taken in conjunction with the accompanying drawings described below.

[0027] FIG. 1 illustrates a block diagram of the direct to digital oximeter as connected to a sensor device; and

[0028] FIG. 2 illustrates the sensor device and direct to digital oximeter connected to a patient.

[0029] A block diagram of a direct to digital oximeter 10 constructed in accordance with the invention, along with an external sensor device 20 is shown in FIG. 1. The direct to digital oximeter 10 comprises a charge digitizing analog to digital converter 30, a microprocessor 40, a digital to analog converter/LED driver 50, and a flash EPROM 60. In order to achieve sufficient accuracy, the charge digitizing analog to digital converter 30 preferably converts the input analog signal to a digital signal of at least 20 bits.

[0030] In a preferred embodiment (see FIG. 2) the sensor 20 is attached to a blood-carrying tissue sample, such as the finger or ear lobe of a patient. Here, the sensor 20 is shown to consist of three light emitting devices 70, 80, and 100 and a single photodetector 90, but the sensor can include two or more light emitting devices of different wavelengths and an associated plurality of photodetectors. Furthermore, although LEDs are commonly used in circuits of this type, the light emitting devices can be laser diodes, white light sources, or other suitable devices. To perform traditional pulse oximetry, however, the sensor 20 comprises a red LED 70 and an infrared LED 80.

[0031] The LED's 70 and 80 are driven by digital signals from the microprocessor 40. These digital signals are converted to analog voltages by means of the digital to analog converter/LED driver 50. Light from the LED's 70 and 80 is transmitted through the tissue sample, and is detected by the photodetector 90, which produces an analog current signal with an amplitude proportional to the amount of light detected in each bandwidth. The current signal from the photodetector 90 is then digitized with 20 bits of resolution by the charge digitizing analog to digital converter 30, and is sent to the microprocessor 40. Demultiplexing, ambient interference identification and elimination, and signal filtering are performed by means of digital signal processing software routines in the microprocessor 40. Once the signals are processed, the microprocessor 40 calculates the value of the ratio

[0032] (IR(AC)/IR(DC))(Red(AC)/Red(DC))

[0033] where the DC component represents the non-pulsating blood flow, and the AC component indicates the pulsatile blood flow. The microprocessor 40 then determines the absolute arterial oxygen saturation by comparing the result to the value stored in a look-up table in flash EPROM 60.

[0034] In one preferred embodiment, the dynamic range of the analog to digital converter is optimized to match the input signal range, thereby providing accurate monitoring over a wide range of input signals. To optimize the dynamic range, the output of the photodetector 90 is initially read to determine the strength of the input signal. If the signal is in a normal or high operating range, a dynamic range control 62 (FIG. 2) is switched into the circuit to reduce the signal amplitude, preventing saturation of the analog to digital converter 30. The dynamic range control 62 preferably comprises a current divider under software control, and is inserted between the photodetector 90 and the analog to digital converter 30. The dynamic range control 62 reduces the current input level by a predetermined amount. In situations where the photodetector 90 output is low (e.g., very dark skin, a thick tissue site, and/or reflectance mode monitoring), the software detects the low digitized signal intensity and switches out the current divider, narrowing the dynamic range and effectively raising the signal gain. While one particular method of providing a dynamic range control is shown, it will be apparent to one of ordinary skill in the art that the dynamic range control could be provided in a number of ways including amplifying a low signal to an expected higher level or reducing the charge integration time of the charge digitizing converter, Furthermore, the dynamic range control may be implemented after current:voltage conversion for a voltage-input analog to digital converter. Additionally, either a fixed or variable reduction might be implemented in the dynamic range control.

[0035] In some applications, it is desirable to add at least one additional wavelength to the sensor 20 to improve the accuracy of the oximetry measurements or to provide additional medical monitoring functions to the oximeter 10. In this case, the light emitting device 100 is added to the sensor 20, and related detector circuitry is added to the photodector 90. Because all signal conversion is time-multiplexed through a single analog to digital converter circuit 30, the only additional circuitry required to add a third or further wavelengths to the oximeter 10 is a driver to drive the additional light emitting devices of the sensor. Preferably, the light emitting devices 70, 80, and 100 are driven by a time-multiplexed digital to analog converter 50, controlled by a software-driven gate. However, it will be apparent to one of ordinary skill in the art that a pulse width modulated (PWM) output could also be used. The applications of the additional wavelength(s) in the sensor 20 include, but are not limited to: noise detection; dyshemoglobin detection and/or measurement; and indicator dye measurement.

[0036] A noise reference signal can facilitate the elimination of noise from a potentially compromised signal source. In pulse oximetry, this can be accomplished by tracking the absorbance of light at a wavelength (e.g., green) which is unaffected by the relative concentrations of different hemoglobin forms, which absorb primarily in the red. The resultant signal is neutral in the absence of noise, but represents fluctuations in intensity due to changes in emitter:detector geometry or other noise sources. The noise reference signal can be employed in a number of known mathematical approaches to noise elimination, including adaptive signal processing.

[0037] Dyshemoglobins occur when the hemoglobin molecule binds with another molecule besides oxygen, and include methemoglobins, sulfhemoglobins, and carboxyhemoglobins. A form which is of particular clinical significance is carboxyhemoglobin, the combination of carbon monoxide with hemoglobin. Carbon monoxide poisoning is a significant cause of morbidity and mortality. Acute cases are often associated-with smoke inhalation at the scene of a fire, but chronic poisoning, wherein a patient presents with “flu” symptoms, may be more insidious. Conventional pulse oximeters cannot readily distinguish carboxyhemoglobin from oxyhemoglobin, resulting in a falsely reassuring oxygen saturation reading. One approach which has been utilized (see U.S. Pat. Nos. 4,167,331, 5,355,880, and 5,413,100, which are hereby incorporated in their entirety) requires a choice of wavelengths near three isobestic points (approximately 580, 650, and 800 nm). However, if detection without exact quantification is sufficient (e.g., to generate a warning), the addition of an 800 nm wavelength (isobestic between reduced and oxygenated hemoglobin) to the basic oximeter is sufficient to identify the presence of carboxyhemoglobin.

[0038] Indicator dyes are introduced as part of several monitoring procedures, including dye dilution cardiac output assessment. For example, U.S. Pat. No. 5,494,031, which is hereby incorporated in its entirety, discloses the use of indocyanine green dye for this purpose, with non-invasive concentration measurement utilizing photoplethysmography. By adding a wavelength of substantially 800 nm to the pulse oximeter sensor, along with known analysis software or methods, and utilizing the infrared wavelength (940 nm) of the pulse oximeter sensor as a reference, a cardiac output assessment function is added to a pulse oximeter.

[0039] Although the oximeter has been described employing three wavelengths, it will be apparent to one of ordinary skill in the art that two or more of the noted features could be added simultaneously by adding additional light emitting devices and associated software to the oximeter.

[0040] In another embodiment of the invention, vectors of infrared and red signal data are stored and used by the microprocessor 40 to determine the arterial oxygenation level. In this embodiment the arterial oxygenation levels are determined as a ratio of observed red ac values (R^obs_ac) to observed infrared ac values (I^obs_ac). A noise metric determined by comparing the observed red signal to a predicted red signal is employed to filter the noise components from the signal, thereby obtaining a more accurate oxygenation reading.

[0041] Following are the steps used to determine the oxygenation saturation level. Assuming for the moment ideal conditions,

[0042] I_dc=LP (I)

[0043] I_ac=I−I_dc

[0044] R_dc=LP (R)

[0045] R_ac=R−R_dc

[0046] where capital letters are employed to indicate vectors of L contiguous data samples (I={i₁, i₂, . . . , i_L}) equally spaced in time with an appropriate sampling rate. Vector length can impact stability of the I:R ratio calculation as well as ability to detect noise in a timely and reliable fashion. The critical timing has been shown experimentally to be the time to slew between minimum and maximum absorbance (caused by the leading edge of the arterial blood bolus), only 100-200 msec in a hemodynamically effective pulse.

[0047] The ac subscripts indicate a high-passed or unbiased pulsatile-component (variation in intensity), and the dc subscripts indicate a low-passed, relatively long-term trend, or bias (the overall intensity level). Here LP() is assumed to have linear phase shift, permitting derivation of the high-passed signal by subtracting the low-passed version from the original. This filtering may be accomplished in hardware, but would be performed by software in the digital oximeter.

[0048] Given R and I, it is known to obtain the SpO₂value by taking the scaled ratio of infrared and red pulse amplitudes, employing an empirically derived proportionality expressed here as an arbitrary function K: ${\mathrm{SpO}}_2=K\left(\frac{\left(\max \left(I_{\mathrm{ac}}\right)-\min \left(I_{\mathrm{ac}}\right)\right)/I_{\mathrm{dc}}}{\left(\max \left(R_{\mathrm{ac}}\right)-\min \left(R_{\mathrm{ac}}\right)\right)/R_{\mathrm{dc}}}\right)$

[0049] where max() and min() denote the signal maxima and minima.

[0050] However, assuming equivalent LED:detector geometry, the I and R vectors are linearly related. One vector, therefore, can be expressed as a simple linear combination (mX+b) of the other. The constant difference is the difference in the low-passed intensity or dc levels, leaving the high-passed components linearly related by the I:R ratio ρ:

[0051] R_ac=ρI_ac

[0052] Then the formula for SpO₂may be rewritten as $\begin{array}{c}{\mathrm{SpO}}_2=K\left(\frac{\left(\max \left(I_{\mathrm{ac}}\right)-\min \left(I_{\mathrm{ac}}\right)\right)/I_{\mathrm{dc}}}{\left(\max \left(\rho I_{\mathrm{ac}}\right)-\min \left(\rho I_{\mathrm{ac}}\right)\right)/R_{\mathrm{dc}}}\right)\\ =K\left(\frac{\left(\max \left(I_{\mathrm{ac}}\right)-\min \left(I_{\mathrm{ac}}\right)\right)/I_{\mathrm{dc}}}{\rho \left(\max (I_{\mathrm{ac}})-\min (I_{\mathrm{ac}})\right)/R_{\mathrm{dc}}}\right)\\ \mathrm{or}\mathrm{just}=K\left(\left(R_{\mathrm{dc}}/I_{\mathrm{dc}}\right)/\rho \right)\end{array}$

[0053] The least-squares minimization (LSM) method is employed to derive p from the signal data by taking

[0054] ρ=(I_ac°R_ac)/(I_ac°I_ac)

[0055] where ° is used here to indicate the dot product of two vectors, yielding in this case the ratio of two scalars. (Note that it is assumed here that the LP() function, although applied continuously over the data, results in unbiased data vectors I_acand R_ac, since the LSM method actually specifies removal of the vector mean from the biased data prior to computation.) If this is not the case, it is possible to recast the calculation of ρ using (I−μ_I) and (R−μ_R) instead and still derive the I:R ratio, assuming appropriate choice of vector length.)

[0056] This calculation of SpO₂is independent of pulse location and peak-valley measurements and will hereafter be called “continuous SpO₂calculation”. However, as will be described below, the calculation still relies upon the fact that pulsatile events takes place within the data vector extent.

[0057] The above calculations were determined under ideal conditions. To obtain an accurate signal, therefore, it is necessary to account for the noise encountered in pulse oximetry calculations. Under non-ideal conditions, the observed intensities are actually

[0058] I^obs=I+N_I

[0059] R^obs=R +N_R

[0060] where N_Iand N_Rare noise components which are assumed to be unbiased (low frequency interference tends not to effect the relatively short data vectors used. An observed I:R ratio ρ_obsis defined by

[0061] R^obs_ac=ρ_obsI^obs_ac

[0062] where ρ_obs=ρ in the absence of noise. Now N_Iand N_Rmay be uncorrelated or correlated, and if the latter, may possess the same or different proportionality ratio as the I:R ratio ρ of the desired signal components I and R. Considering the most difficult situation of correlated noise, noise possessing the same ratio as ρ would not effect the LSM calculation of ρ_obsfrom I^obsand R^obs, giving ρ=ρobs. However, assuming a general case of

[0063] N_R=_ρNN_I

[0064] with ρN≠ρ, the noise components must be removed or canceled to accurately calculate ρ_obsand thence SpO₂directly from I_obsand R_obs.

[0065] Since the SPO₂value represents the binding state of millions of hemoglobin molecules, as determined by relatively slow processes such as alveolar transport of molecular oxygen, pumping of blood through the circulatory system, and venous return through the capillary beds, the pulse-to-pulse variation in the SpO₂level is relatively small.

[0066] Motion artifact, however, tends to appear rather suddenly, induces non-linear effects on the sample-to-sample relationship between R^obs_acand I^obs_ac, and disturbs the observed I:R ratio. Noise attributed to motion artifact, therefore, must be filtered to obtain an accurate calculation. The noise can be quantified with a noise metric between R^obs_acand the predicted red signal, obtained from I^obs_acby assuming a constant estimated _ρest(derived from the recent history of ρ_obsunder low noise conditions):

[0067] RP^red_ac=ρest^Iobs_ac

[0068] Letting Δ() indicate a desired noise metric,

[0069] ν=Δ(R^obs_ac, RP^red_ac)/α

[0070] where α is a normalization factor (required if the metric is not inherently normalized). The metric is defined so that ν is zero only if ρ_obs=ρ, and otherwise is positive and increasing with increasing disturbance of either or both observed intensity signals.

[0071] One distance metric is the average absolute difference between corresponding vector elements, or $\Delta \left(V,W\right)=\left(\sum _{j=1}^L\left|v_j-w_j\right|\right)/L$

[0072] which is simple to compute and may be recursively obtained. This metric, however, requires a normalization for the expected magnitude of the signals. This normalization must be obtained from I^obs_acand R^obs_acdata gathered under low noise conditions (denoted R^est_acand I^est_ac). One example is the maximum magnitude of the two estimated signals, or $\alpha =\left(\sum _{j=1}^L\max \left(\left|r_j\right|,\left|i_j\right|\right)\right)/L$

[0073] It will be apparent to one of ordinary skill in the art that other distance metrics (including correlation) can be applied to the problem of comparing R^obs_acto RP^red_ac.

[0074] By empirically establishing a threshold for acceptable performance, the noise metric ν, calculated at the same time as the new ρ_obs, can be used to control not only SpO₂averaging but other pulse oximetry processing (such as pulse rate determination).

[0075] As pointed out above, the continuous SpO2 calculation does not require determination of pulse timing. However, since it is desirable to limit the vector length (e.g., to less than one second), it will be possible in low heart rate situations to obtain data vectors containing no pulsatile event. These vectors will show less of the arterial absorbance effect which is the basis of pulse oximetry. In preferred embodiments, therefore, an ECG or other indicator is used to synchronize the data collection.

[0076] While preferred embodiments of the invention have been shown and described, it will be clear to those skilled in the art that various changes and modifications can be made without departing from the invention in its broader aspects as set forth in the claims provided hereinafter.