Efficiency of Hair Detection in Hair-to-Hair Matched Trichoscopy

Introduction Precise evaluation of changes in hair count is crucial for monitoring progression of hair loss and the effects of treatment. The focus of this study is the comparison of the various examination and assessment techniques in terms of the precision of hair count change observed in trichoscopy images. Methods Controlled hair extraction of the same scalp spot was used to simulate hair loss, and the different examination techniques were performed to detect this change. The investigators who performed the counting were blinded. Results For trichoscopy images, the average error in determining the terminal hair count change (relative to total hair count) was 9 ± 1% for automatic assessment with manual correction and 0.4 ± 0.2% for hair-to-hair matched images. For phototrichogram, the automatic measurement results were found to deviate from truth on average by 12 ± 2%. The manually corrected hair count results were much closer to the truth with average deviation at the level of 7 ± 1%. The hair-to-hair matched results corresponded to approximately 0.6 ± 0.3% average discrepancy. Conclusion Combination of manually corrected image processing, follicular mapping, and hair-to-hair matching appears to be the most precise way of evaluating the change in hair count over time. These novel techniques should be considered valuable, especially in research and clinical trials.


Introduction
Precise evaluation of changes in hair count are crucial for monitoring progression of hair loss and the effects of treatment [1][2][3][4]. Automated software tools, manual hair counting (also called manually corrected count), and hair-to-hair (H2H) matching can all be used to statistically process microscopic hair images [5,6]. Although some of these methods are known to measure hair count only approximately, they are believed to correctly indicate its change in before and after comparisons [7][8][9].
The focus of this study is the comparison of the different examination techniques in terms of the precision of the before and after hair count change. Only terminal hair shafts (thickness >40 μm) were taken into account as they form the majority of hair coverage and volume studied in most clinical and research projects. This article is licensed under the Creative Commons Attribution 4.0 International License (CC BY) (http://www.karger.com/Services/ OpenAccessLicense). Usage, derivative works and distribution are permitted provided that proper credit is given to the author and the original publisher.

Materials and Methods
Healthy volunteers with contrasting hair color were selected: five from among the staff of Grimalt Dermatology, Barcelona, Spain, and four from among the staff of Sinclair Dermatology, Melbourne, Australia. The group comprised 5 females and 4 males aged 17-45 years. All subjects gave their oral consent to participate in the study in the presence of at least one witness. The subjects washed their hair the same day to remove hair that could accidentally fall out during the examinations. The different examination techniques were applied one after another on exactly the same test spot on patient's scalp. To determine the sensitivity of the techniques to hair count change, the following experiment was performed. Controlled hair extraction between subsequent examinations of the same spot was used to simulate hair loss, and the different examination techniques were performed to detect that change. In the cases of examination techniques involving human experts, a blinding procedure was used. The details of controlled hair extraction were known only to the examination team and were not disclosed to the laboratory staff responsible for image analysis, manual hair counting, or H2H matching. Test Procedure Trichoscopy 1. A single examination spot was selected and marked with two felt-tip pen dots ca 1 cm apart, as illustrated in Figure 1. The same test spot was used throughout the whole examination procedure. 2. First set of trichoscopy examinations -standard trichoscopy examination and H2H matching procedures -were performed (3 sequential images of the same spot with complete rearrangement and re-parting of the hair between taking images).
3. First controlled hair removal using high magnification glasses and metal tweezers, the investigator carefully removed a few hair shafts from the test spot and made photographic documentation of the removed hair shafts and their location in relation to the two spot marking. 4. Second set of trichoscopy examinations -complete standard trichoscopy and H2H matched trichoscopy -were repeated in the same way as before the hair removal. Phototrichogram 1. Hair clipping -hair from the area approximately 1.5 cm 2 around the test spot marks was clipped to ca 1 mm and the hair remnants carefully removed. 2. First set of PTG examinations -the first image of the test spot was recorded with DermoGenius for the DermoScan analysis; subsequent images were recorded with FotoFinder medicam both for manual processing and the H2H matching PTG procedure (3 images of the same spot taken after combing of the stubble in different directions to rearrange it) [10]. 3. Second controlled hair removal -the investigator removed further hair shafts from the test spot and made photographic documentation of the removed hair shafts and their location in relation to the test marks. 4. Second set of PTG examinations -the complete set of PTG images for the procedures under study were repeated in exactly the same way as before the controlled hair removal. The PTG and the trichoscopy images obtained before and after controlled hair removal were submitted to TrichoLAB for processing. The dates/times of randomly selected examination files were modified so the lab staff could not know which was the initial one and whether to expect hair loss or hair gain. The details of controlled hair removal for each patient were known only to the examiners at Grimalt Dermatology and Sinclair Dermatology. They were disclosed only for final evaluation of examination results.
The examinations were performed at Grimalt Dermatology, Barcelona, Spain between February 20 and 26, 2019 and at Sinclair Dermatology, Melbourne, Australia between December 2, 2020 and January 20, 2021. As only 5 out of the 9 volunteer subjects agreed for hair clipping, only these subjects took part in the PTG efficiency measurement. For trichoscopy efficiency measurement, the data of 1 subject was excluded due to poor image quality.  Table 1 presents the results of trichoscopy examinations: the manually corrected hair counts (columns 1&2) and the H2H matching results (columns 3&4) in baseline and follow-up examinations. The change in hair count derived with these two techniques is presented in columns 5 and 6, respectively. Figure 2a, b present the H2H matched trichoscopy images of the right dot mark of subject no. 1. The two green-labelled hair shafts in Figure 2b have been detected as new hair shafts that were not pres-

The Trichoscopy Measurement Results versus the Truth
Once the trichoscopy examination processing was complete, the unblinding procedure was performed and column 7 was added to Table 1. It presents the number of terminal hair shafts removed in the controlled hair extraction procedure with a plus or minus sign depending on whether the baseline and follow-up examinations were swapped or not. Figure 3 shows documentation of four hair shafts extracted from subject no. 1 between the two trichoscopy examinations. As the dates of the first and the second examinations were swapped, results are presented in column 7 as a gain of four new hair shafts.
Columns 8 and 9 present the extent to which the results obtained with the two trichoscopy techniques deviate from the true change in terminal hair count. The manually corrected hair count change was found to deviate from the truth on average by 9 ± 1% (relative to the total number of measured hair). The H2H matched results were wrong in 6 cases out of 1,382 measured hair shafts, corresponding to 0.4 ± 0.2%. Table 2 presents the results of PTG examinations: automatic (columns 1&2) and manually corrected hair counts (columns 3&4) in before and after examinations as well as the H2H Matching results (columns 5&6). The change in hair count derived with these three techniques is presented in columns 7, 8, and 9, respectively. Figure 4 presents an example of a TrichoScan TM report for subject no. 1; because the standard report presents only the density of terminal hair, it was multiplied by the measurement area for the comparison. Figure 5a, b present the H2H matched PTG images for subject no. 1. Only one new hair was detected in Figure 5b (labelled green) that was not present in Figure 5a, as noted in column 9 of Table 2 accordingly.

The PTG Measurement Results versus the Truth
After completing the PTG examination processing, the unblinding procedure was performed and column 10 was added to Table 2. It presents the number of terminal hair shafts removed in the controlled hair extraction procedure between the two PTG examinations with a plus or minus sign depending on whether the two examinations were  swapped. As presented in Figure 6, only two clipped hair shafts were extracted from subject no. 1, and the two PTG examinations were swapped. The true change presented in column 10 for patient 1 is therefore +2 hair shafts. from truth on average by 12 ± 2%. The manually corrected hair count results were much closer to the truth as the average deviation for the given subject sample was on the level of 7 ± 1%. The H2H matched results were wrong in 4 cases out of 724 measured hair shafts, corresponding to 0.6 ± 0.3% discrepancy.

The Precision of the Results
The resulting detection inefficiencies are quoted with the estimation of statistical precision (after ±) derived, assuming that the number of hair shaft misdetections follows Poisson distribution. These estimates were precise enough for the purpose of this study, so it was concluded that the number of subjects was also sufficient. Analysis of systematic uncertainties generally indicated that the denser the hair patient has, the higher the multiplicities of follicular units, the more difficult it is to detect them and the larger the ineffi-ciencies are. To address this issue, young and healthy individuals with dense hair were recruited for the study.

Conclusion
The manually corrected PTG results are approximately twice as accurate as the automatic assessment of terminal hair count difference between the baseline and the follow-up examinations. Although more precise and reliable, manually corrected PTG suffers from similar limitations: hard to detect hair shafts sticking closely together in tight follicular units and slightly different field of measurement in the before and after examinations. The effect that was observed, but could not be studied further, is the dependency of the automatic PTG result on the clipping length; in our study, the subsequent examinations were performed with exactly the same clipping.
The H2H matched analysis of both the PTG and the trichoscopy examinations is over one order of magnitude more accurate in terms of before and after comparison. The reasons for this fact are as follows: • The F-Mapping procedure allows recovering information about hair that could not be detected in the primary image (e.g., because their view was obstructed by other hair) from the subsequent two images of the same spot with a different hair arrangement. The F-Mapping procedure brings the effective hair detection efficiency from 90 to 95% to over 99.5%. • Before and after matching procedure ensures that precisely the same scalp area is used for the comparison, compensating for different camera positioning as well as any skin stretching/distortions.  Automatic analysis using TrichoScan TM software provides quick but precision-limited results -the measured hair count change was found to deviate from truth on average by ± 12% and the deviation could increase if the clipping in the before and after examinations was not identical. The manually corrected PTG and trichoscopy were determined to have fewer errors per 100 measured hair shafts (7% and 9% for PTG and trichoscopy, respectively) and are therefore advocated as the technique of choice when high precision is demanded, e.g., in clinical studies. The H2H matching further improves precision of the before and after comparison, reducing the number of errors to 0.4% and 0.6% for trichoscopy and PTG, respectively.
Combination of manually corrected image processing, follicular mapping, and H2H matching appears to be the most precise way of evaluating the change of hair count over time in clinical trials and hair research. Currently, as this combined procedure is time-consuming, its everyday use is rather limited. Hopefully, future developments should make this technique more available for office use.

Statement of Ethics
All the subjects recruited from among medical staff gave their written consent to participate in this study. Ethical approval was not obtained for this study as it was not needed as per national guidelines.