• info@howismyhearing.net

Category ArchiveA.I.

Audiogram Data Points

Standard Audiological Pure Tone Audiogram                                                                                                                                                

Has 7 Frequencies

250 to 8000: 5 octave intervals and 2 half octave intervals

Has 21 decibel levels per frequency

5dB increments 0-100 for each frequency

NUMBER OF DATA POINTS ON AN AUDIOGRAM

144 DATA POINTS

Number of Combinations for pure tones

5,786,051*

*The combination of 144 decibel levels in 5 dB increments and 7 frequencies is a generally accepted professional standard for PATTERN AND LEVEL data points used to classify pure tone audiograms that are used to determine next steps leading to a hearing diagnosis.

The 144 data points do not include the data points needed for bone conduction, SRT & PB speech tests, false positive and false negative values, otological confirmation which makes up another 150 data points that were included to ensure polynomial / pattern recognition accuracy (confidence coefficients) Inclusion of these values increased the number of combinations to over 10 million for audiogram classification

pure tone audiogram

pure tone audiogram

This document I put together is to point out the vast number of data sets and complexity of the training data required to develop my pattern recognition system. It only addresses the most basic audiometric tests that were needed to ensure it was both an audio-logically and medically accurate interpretation of a pure tone audiogram.

The polynomials need to accurately reflect the pattern and level of a pure tone audiogram for all age ranges, genders and healthcare markets.  

 

Hopefully this document will provide some awareness of the vast amount of data points needed to develop the training data.

It should also serve to answer the question people have asked me repeatedly since 1997 “why can’t anyone with software and some knowledge of hearing duplicate what you did, it seems simple to do there are only 4 audiogram patterns”.

Since 997 many millions of tests have been carried out and classified using the system with the embedded polynomials. These markets included one time or another, industrial hearing conservation programs, schools, audiological practices, medical practices, pharmacies, retail hearing aid dispensers, military hearing programs, hearing aid manufacturing businesses, audiological hardware manufactures.

At no point in these millions of tests did anyone challenge the accuracy of the pure tone system. The most common resistance came from the displacement of human involvement in the actual conducting of the manual pure tone test and its interpretation. Now that software disintermediation is more common place its everyday use will have less resistance and enable consumers to directly make better hearing healthcare decisions.    

 

The initial patient training data involved over 1500 patient samples derived actual patient testing using a manual clinical audiometer and 2 types of automated computerized pure tone testing hardware.

The initial patient data needed to be expanded to over 5000 in order to account for patient outlier’s and improve confidence coefficients of the polynomials.   Even with the expansion to 5000 I still had outlier’s, so we developed an audiological programable random audiogram generator to deal with the remaining outliers. The RAG was necessary because the remaining outliers were so rarely seen in a clinical practice here was not enough real patient data to retrain the system to improve the confidence coefficients. Most of the rarely seen outliers were in found in practices that combined audiological / otological medical practices, but the numbers were not sufficient to retrain the polynomials accurately. The other difficulty in data collection was the necessity for inclusion all age ranges, infants to old age, along with sex and a medical confirmation, as all five hearing testing markets had unique and specific methods of testing and data collection based the hearing program goals and reported outcomes of the testing program. Many early pure tone hearing testing program goals did not yield the expected outcome and were abandoned because of conflicting interests and costs / ROI.

Errol Davis   Audiologists ©