An algorithm for automated extraction of pressure and timing data from HRM spatiotemporal plots is presented. Strong correlations were observed between data extracted using automated and manual methods, demonstrating the accuracy of the algorithm. Generally, pressure values were computed more consistently than timing values. Extracting timing measurements is more subjective than pressure measurements, as pressure rise onset and pressure fall offset can be slightly ambiguous. The lowest correlation coefficients recorded were for tongue base pressure rise rate in normal subjects and velopharynx pressure rise rate in disordered subjects; however, results from the two methods were still highly linked with correlation coefficients of 0.869 and 0.825, respectively. Because an automated algorithm eliminates the subjectivity inherent to manual extraction, implementation could lead to more consistent measurements. Optional user intervention, while needed rarely in this study, ensures that the correct sensors are analyzed. Intervention occurrence is dependent on factors such as the nature of particular swallowing disorders and also erroneous sensor readings. For example, the data trace of one subject was marked by several periods of a false high pressure band which caused errors in the detection of the velopharyngeal region. Overall, the rate of user intervention in this study was approximately 10%. Both automated and manual analyses are subject to biases and limitations. Automated analysis is limited by our descriptions of how complex physiological phenomena may appear and by a potential inability to operate predictably when faced with an extremely disordered swallow. Manual analysis is limited by the subjectivity and potential inter-rater reliability issues which plague all methods of manual data interpretation and extraction. Such biases cannot be controlled and are not consistent across raters; however, automated analysis uses the same process every time and thus may be more reliable.
Automated analysis also provides the opportunity to evaluate parameters which cannot be measured manually, such as the area and line integrals evaluated in this study. Area integrals were rather stable within subjects compared to the intrasubject variability of maximum pressure measurements. Total pressure generated is likely more physiologically meaningful than maximum pressure, reflective of the overall bolus propulsion force. Measurement of pressure maximums is subject to including pressure spikes of short duration which may not significantly contribute to bolus transit. Integrals provide a more comprehensive manometric picture that is immune to such phenomena. Poor clinical decisions could potentially be made if based solely on local maximum pressure data. For example, a tongue base pressure curve of a patient with muscle weakness may mask the deficit present and make the patient appear normal. Conversely, a pressure spike in a healthy patient with otherwise normal pressure generation may cause a clinician to diagnose the patient with muscle hypercontractility.
Line integrals may be valuable in evaluating patients who can generate, but not sustain pressures either at the velopharynx or the tongue base. This would produce a large value for the line integral, but small value for the area integral (). Partial area integrals are also useful in characterizing the shape of a pressure curve. If near total pressure generated is reached at approximately 75% of maximum pressure (), a pressure spike may be skewing maximum pressure measurement while not generating meaningful physiological pressure for swallowing. Though not observed in this study, one could expect differences in pressure wave velocity between normal subjects and patients with disorders affecting pressure propagation. Modifying and combining the novel parameters measured in this study into potentially clinically useful parameters and ratios will be the subject of future investigations.
Figure 6 Pressure traces from the tongue base of three different subjects. (A) Large value for both line and area integrals. (B) Large line integral with small area integral. (C) Pressure spike occurring at end of curve which has little impact on total pressure (more ...)
Pressure gradients represent the force underlying bolus propulsion. We observed that a large negative gradient immediately preceding UES opening became a large positive gradient at the time of maximal UES opening, indicated by minimum UES pressure. The three disordered subjects in this study displayed different irregularities in this pattern. For example, subject 1 has a positive gradient at maximum UES pre-swallow pressure, possibly indicative of high velopharyngeal activity trying to compensate for UES dysfunction. Data from subjects 2 and 3 were closer to those from normal subjects; however, subject 2 may have cricopharyngeal hyperactivity due to the large negative gradient and smaller positive gradients. Subject 3 may have slight UES hypofunction, due to the reversal of the trend seen in subject 2, relative to normal subjects.
Three disordered subjects were analyzed in this study. No modifications were made to the algorithm to accommodate analysis of these subjects. The few problems which occurred were problems occasionally experienced when analyzing normal subjects as well. Pre- and post-swallow UES pressure maximums were lower than the average of the normal subjects, but were typically within a range seen in normal subjects with lower pressures. In some situations, the lower pressures required user intervention to correct the suggestions of the program. Maximum tongue base pressure remains the most difficult to locate, for both normal and disordered subjects. This may be a consequence of the simplicity of our detection algorithm; further refinements will focus on improving performance while not sacrificing ease of use or processing speed. No formal statistical analyses were performed to determine the presence of differences between normal and disordered swallowing, as we tested a limited number of disordered subjects and such a comparison is beyond the scope of this study. Several observable differences for parameters such as velopharyngeal pressure rise rate, velopharyngeal pressure integral, maximum UES pre-swallow pressure, minimum UES pressure, and total swallow duration will be the subject of future investigations. One interesting feature found in the spatiotemporal plots of normal subjects which appears absent in those from the disordered subjects is a linear pattern connecting the pressure peaks within the swallow. While pressure peaks created by the velopharynx, tongue base, and UES align linearly in normal swallows, this linearity is lost in disordered swallows (). This may be caused by cricopharyngeal dysfunction. Hyperfunction could result in a common cavity event, with uniform pressure from the velopharynx to tongue base. Prolonged UES opening could also disturb the linear pattern. While we tested a variety of types of dysphagia to evaluate the robustness of our automated analysis algorithm, systematic comparisons between normal and disordered subjects may benefit from standardizing the type of dysphagia present.
Figure 2 Spatiotemporal plots from three disordered subjects. Subject 1 (left) displayed high velopharyngeal pressure (A), a short duration of tongue base pressure and small tongue base pressure integral (B), low pre-swallow upper esophageal sphincter (UES) pressure (more ...)
To extend automated analysis to patients with more severe dysphagia and more atypical pressure patterns, our algorithm may need to be altered. The algorithm currently locates the UES first. Patients with cricopharyngeal hypofunction may have low resting UES pressures or display different UES activity during swallowing. As UES pre- and post-swallow pressure maximums are used to calculate timing events, both timing and pressure parameters may be affected. Also, although our algorithm is still effective at relatively low pressure values, patients with severely reduced tongue base retraction may be unable to generate a great enough pressure differential to be recognized. Solving this problem could be done either by modifying the current algorithm, or by creating a set of disorder-specific algorithms which could be applied when appropriate. At this time, applying the algorithm and relying on user intervention to correct program suggestions when necessary, particularly in the tongue base region, could be employed. While the automated algorithm shows promise, this is a preliminary study. The ability to analyze data obtained from a wide variety of clinical patients must be demonstrated before the algorithm can be applied clinically.
The large amount of data generated by HRM is well-suited for automated analysis. While one may be able to look at a swallow and perceptually determine that it is abnormal, determining which treatment is appropriate demands knowing how that swallow is abnormal. The potential number of variables which can be analyzed is infinite and automated analysis is needed if data extraction is to remain feasible. As more complex disorders are evaluated, the absence of certain key events (e.g. tongue base pressure peak) may become as significant as their presence in normal subjects. Further, additional parameters such as gradients, integrals, and pressure wave velocity can be calculated with automated analysis and may provide a more robust picture of overall swallowing patterns. Though not specifically measured, the time required for the automated analysis was much shorter than for a manual extraction, even without the additional parameters. While manually extracting all data from a single swallow takes approximately three minutes, it can be done in a few seconds using our automated algorithm. In the case of user intervention to correct sensor identification, only an additional 15–20 seconds are required, though this time is dependent on the experience level of the user. Automation also allows for more precise pressure measurements, as the possibility of users selecting an incorrect maximum is removed. The efficiency and accuracy of our algorithm make it a potentially valuable tool if HRM is to be applied routinely to patient assessment.