12 technical documentation

Sensor technical information

brain4care reserves the right to limit the provision of diagrams, bills of materials, drawings and other documents related to the construction of the product. This type of information is restricted to authorized service personnel only.

Analytical software technical information

The B4C System analytical software produces information output such as estimated P2/P1 ratio variation over time, surrogate ICP waveform signal over time, pulse morphology as average per minute with estimated parameters of P2/P1 ratio, number of useful pulses, heart rate (bpm), and normalized time to peak. 

Calculations are done via processing algorithms that receive raw data from the B4C System sensor and generates a processed PDF report with the surrogate ICP waveform and estimated associated waveform parameters.

Average pulse calculation

The automated analysis, done for each minute at a time, includes all steps briefly described below which were developed based on established principles and methods described in literature, to isolate the desired surrogate ICP waveform signal for consistent processing:

• Data input - data from a monitoring session that is temporarily stored in the Mobile App, which is connected to the B4C System sensor, is transferred to the B4C System. De-identified data is parsed  and saved for processing. 

• Downsampling - the acquired signal is downsampled to a desired sampling frequency for analysis.

• Signal de-trending - given the device’s mechanical properties certain trends in the monitoring session timeline are eliminated. 

• Signal filtering - high frequency waveform noise is removed.

• Signal validation - non physiological waveforms are identified and removed.

• Pulse finding - pulses contained in the captured signal are identified. At this point the subject device’s output derived heart rate (bpm) is calculated.

• Check inversion - mirrored waveforms are identified and inverted.

• Artifact removal - patient head movements, as well as bites, cringes, etc… are captured by the sensor and may create waveform pulse artifacts. These are identified and removed. At this point the system calculates the number of useful pulses, i.e., the number of pulses considered in the calculation of the average pulse.

• Pulse alignment - after all pulses have been validated, they are symmetrically padded , i.e, they are “stretched” or “shrunk” time-wise in order to normalize their lengths without distortion.

• Pulse averaging - all the aligned useful pulses for that same minute are finally used to calculate the mean pulse with a 95% two-tailed confidence interval. The mean pulse is the visible blue line and the grey zone is the confidence interval as seen in figure below. At this point the subject device’s output pulse morphology as average per minute is generated. 

Open

processed pulse morphology for that minute

The analytical software up to this point transformed the raw data into an average pulse waveform a minute at a time. As a monitoring session progresses, the analytical software uses these series of verified mathematical methods (described above) applied in the medical software. These verification methods provide quality checks to ensure the analysis is performed correctly and consistently on valid signals. 

For each minute of average surrogate ICP waveform, two key parameters are derived from relevant physiological information based on relative and / or normalized measurements. The most relevant parameters meeting those conditions are estimated P2/P1 ratio and normalized time to peak and they are described as follows. 

Estimated P2/P1 ratio

P2/P1 ratio’s key hypothesis is that when the P2 relative amplitude is higher than P1 it suggests a loss of compensatory reserve mechanisms and deterioration of intracranial compliance where P2 tends to increase if the compliance is reduced. The simple observation of ICP peaks by the physician may lead to misjudgment, since P2 may be overlapped with the respiratory cycle, given a false impression of elevation over P1. Hence, the automated estimation of P2/P1 in accordance with the cardiac cycle described below may provide more accurate information.

An algorithm was developed based on a hypothesis derived from the observation of the data from clinical trials. Data supported the observation that the percussion and tidal waves of an ICP pulse (P1 and P2 respectively)  are not always visible given the combination of complex intracranial mechanisms. Albeit the mechanisms of percussion and tidal wave do occur inside a patient's skull, the resulting visual representation for them might not always be apparent for a naked eye. As such, a method to find a tidal wave (P2) albeit not visually identifiable was developed.

Based on the the observation of a subset of 1453 minutes of concurrent invasive ABP (arterial blood pressure) and invasive ICP (intracranial pressure) as well as non-invasive BcSs-PICNI-2000 (B4C Sensor device) from a clinical study demonstrated a significant statistical correlation between the waveform parameters shown on figure below:

• Invasive ICP’s percussion wave (A1) timing (T1) is almost the corresponding ABP’s waveform first peak 

• Invasive ICP’s notch (A3) timing (T3) is closely related to ABP’s dicrotic notch

• Invasive ICP’s tidal wave (A2) timing (T2) is estimated to be midway between ABP’s T3 and T1

Open

Representation of observed correlation between arterial blood pressure ABP (red) waveform and intracranial ICP (green) waveform. ICP’s percussion wave lag (A1) timing (T1) in the ICP pulse is almost the same as ABP’s first peak; ICP’s notch (A3) timing (T3) after its percussion wave is closely related to ABP’s dicrotic notch; ICP’s estimated tidal wave (A2) timing (T2) is observed to be midway between ABP’s T3 and T1

ABP based algorithm (research algorithm)

Based on this observation an algorithm was developed that estimates the timing of P1 and P2 amplitudes for the invasive ICP waveform, as well as the BcSs-PICNI-2000 (B4C Sensor) waveforms, based on the concurrent ABP pulse’s T1 and T2 normalized timings. This allowed for indirect (surrogate) P1 and P2 estimation when they are not visually identifiable in the invasive ICP.

The accuracy of this ABP based P2/P1 ratio estimation algorithm (ABP based algorithm) was validated through a resulting Pearson correlation of 0.845 [0.797, 0.895], p-value 0.076 [0.000, 0.932]f, and normalized mutual information correlation of 0.772 [0.647, 0.923]

Estimated P2P1 ratio algorithm (production algorithm)

After verification of the ABP based algorithm (research algorithm), a mathematical model was developed to allow for P2/P1 ratio estimation without depending on having a concurrent ABP pulse waveform information being captured by the B4C System’s sensor since it is not the actual usage scenario. A production algorithm was developed using available pulses of invasive ICP (invasive device) and non-invasive BcSs-PICNI-2000 (B4C Sensor) sensors that had their P1 and P2 amplitudes calculated by the ABP based algorithm. With these peaks identified based on the corresponding ABP pulse, the normalized timing of the amplitudes (moment during the pulse’s duration where the peak happens) were defined from the max slope (moment in time where pulse slope is steepest) to the timing of identified P1 and P2 peaks and then normalized by the pulse total time. This then determined the most likely positions of P1 and P2 amplitudes.

A subsequent test was done to validate the production algorithm. All valid waveform data captured from the invasive ICP (invasive sensor) and non-invasive BcSs-PICNI-2000 (B4C Sensor device) had their minute by minute P2/P1 ratio calculated with the ABP based algorithm (research algorithm) and Production algorithm. When the P2/P1 ratio was calculated by both algorithms, one being the P2/P1 ratio estimated by using the concurrent APB waveform as defined in figure above, and the other being the estimation of the most probable position of P1 and P2, and then compared to each other, the correlation was high between both methods when applied to the waveform coming from both sensors:

• invasive ICP waveform: Pearson correlation of 0.916 [0.902, 0.930], p-value 0.000 [0.000, 0.000]

• non-invasive B4C Sensor waveform: Pearson correlation of 0.867 [0.844, 0.887], p-value 0.000 [0.000, 0.000]

Normalized time to peak

Normalized time to peak’s key hypothesis is that it is affected by compensatory reserve and intracranial compliance. The derived parameters required to identify time to peak are defined in figure below. Normalized time to peak is identified from point A, the max slope of the calculated average waveform (moment in time where pulse slope is steepest) to point B, normalized duration from the identified average pulse’s Max slope to highest pulse amplitude.

This parameter helps estimate when the highest peak of a pulse happened, normalized to the pulse's total time. The hypothesis is that if the time to peak is at a later duration of the pulse’s normalized timeline, it suggests being the waveform’s tidal wave P2. If the time to peak is at an earlier moment in the pulse’s normalized timeline, it suggests being the waveform’s  percussion wave P1. Such interpretation should always be done by a trained health professional.