Detection of Muscle Activation during Resistance Training Using Infrared Thermal Imaging
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
sensors Article Detection of Muscle Activation during Resistance Training Using Infrared Thermal Imaging Haemin Jung 1,† , Jeongwung Seo 2,† , Kangwon Seo 2,† , Dohwi Kim 3 and Suhyun Park 1, * 1 Department of Electronic and Electrical Engineering, Ewha Womans University, Seoul 03760, Korea; goalssla12@cau.ac.kr 2 School of Electrical and Electronics Engineering, Chung-Ang University, Seoul 06974, Korea; cyh4681@cau.ac.kr (J.S.); skangwon@naver.com (K.S.) 3 Thermoeye, Seoul 06979, Korea; oranss@thermoeye.co.kr * Correspondence: suhyun.park@ewha.ac.kr † These authors contributed to this work equally. Abstract: Infrared thermal imaging has been widely used to show the correlation between thermal characteristics of the body and muscle activation. This study aims to investigate a method using thermal imaging to visualize and differentiate target muscles during resistance training. Thermal images were acquired to monitor three target muscles (i.e., biceps brachii, triceps brachii, and deltoid muscle) in the brachium while varying the training weight, duration, and order of training. The ac- quired thermal images were segmented and converted to heat maps. By generating difference heat maps from pairs of heat maps during training, the target muscles were clearly visualized, with an average temperature difference of 0.86 ◦ C. It was observed that training order had no significant effect on skin surface temperature. The difference heat maps were also used to train a convolutional neural network (CNN) to show the feasibility of target muscle classification, with an accuracy of 92.3%. This study demonstrated that infrared thermal imaging could be effectively utilized to locate Citation: Jung, H.; Seo, J.; Seo, K.; and differentiate target muscle activation during resistance training. Kim, D.; Park, S. Detection of Muscle Activation during Resistance Training Keywords: infrared thermal imaging; muscle activation; skin temperature; deep learning; brachium Using Infrared Thermal Imaging. Sensors 2021, 21, 4505. https:// doi.org/10.3390/s21134505 1. Introduction Academic Editor: Paola Saccomandi Resistance training exercises cause a subject’s muscles to contract against an external resistance, which eventually strengthens and builds muscle endurance. Manipulating Received: 4 June 2021 training variables (e.g., load weight, recovery period, and lifting speed) can produce a wide Accepted: 26 June 2021 range of training stimuli [1]. Thus, various strategies have been explored to quantify the Published: 30 June 2021 effects of resistance training. Marston et al. proposed a metric called exercise density and investigated its relation to a physiological marker of internal training intensity along with Publisher’s Note: MDPI stays neutral traditional measures (e.g., number of repetitions and volume load) of external training with regard to jurisdictional claims in intensity [2]. Another study calculated mechanical work to quantify external resistance published maps and institutional affil- training volume during a resistance training session [3]; however, this process is time iations. consuming and requires specialized equipment, thereby limiting its practical application [4]. Although quantitative amounts of training can be measured, the direct effects of training on the muscles are not considered. Surface electromyography (sEMG) has been widely used to quantify muscle effort during resistance exercises via direct monitoring of muscle Copyright: © 2021 by the authors. activation [5]. By evaluating the zero-crossing levels of sEMG signals, assessments of a Licensee MDPI, Basel, Switzerland. subject’s fatigue endurance have been presented [6]. However, sEMG requires placement This article is an open access article of surface electrodes on the target muscles, which impede the subject’s movements during distributed under the terms and training. In addition, there are various factors that affect impedance measurements between conditions of the Creative Commons Attribution (CC BY) license (https:// muscles and sEMG electrodes, such as biological tissue composition, distance between creativecommons.org/licenses/by/ muscle and electrodes, and size of the electrodes [7,8]. 4.0/). Sensors 2021, 21, 4505. https://doi.org/10.3390/s21134505 https://www.mdpi.com/journal/sensors
Sensors 2021, 21, 4505 2 of 12 During resistance training, heat is dissipated from muscle contraction owing to metabolism and blood perfusion. Therefore, some studies have investigated the rela- tionship between temperature changes and training [9–14]. A previous study showed that prolonged exercise increased the amount of heat dissipated from muscle [9]. Several works have utilized thermistors to investigate changes in the local temperature of muscles after exercise [10]; however, these measurements were invasive, as the thermistors were inserted into the subjects’ veins. Owing to its noninvasive and noncontact nature, infrared thermal imaging has been widely used for various biomechanical applications [11–13]. Using infrared thermal imaging, several studies have investigated skin surface temperature changes related to muscle activation [9,15–19]. It is known that human skeletal muscles have a mechanical efficiency of 30–65%, and there is heat dissipation (i.e., temperature change) related to muscle contraction [9]. Zagrodny et al. showed that there is a correlation between sEMG and thermal characteristics of the skin surface [15]. Daud et al. also showed a correlation between sEMG signals and skin temperature measured from thermal imag- ing [16]. In [17], the authors observed that the temperature of the quadriceps before and after training showed substantial agreement with changes in skin temperature following physical exercise. Another study investigated localized heating of limbs in individuals of different age groups [18]. Furthermore, temperature changes from high-intensity exercise associated with muscle fatigue have also been investigated [18,19]. Therefore, previous studies have proven that the change of skin surface temperature measured from thermal imaging is highly correlated with the isolation of muscle activation. Although the cor- relation between muscle activation and thermal characteristics has been demonstrated in some of the previous studies, these were limited to a single region or muscle. Thus, it was necessary to investigate a method to locate and differentiate muscle activation during training. This study aimed to investigate the use of infrared thermal imaging to assess the isolation of muscle activation during resistance training. Based on previous studies, it was assumed that the skin surface temperature measured from thermal imaging can be directly related to muscle activation. Thermal images were acquired during resistance training while varying the training conditions. The thermal images were then processed and analyzed to visualize muscle activation. To evaluate the performance of the proposed method, a deep-learning model was trained and tested on the processed thermal images for classification of the target muscle activation. The advantage of this study is that the activations of different muscles can be located and differentiated by utilizing the thermal images acquired in a noninvasive and noncontact manner. 2. Materials and Methods 2.1. Experimental Setup Figure 1 shows the experimental setup used to detect muscle activation. Thermal images were acquired using a thermal camera (A65SC, FLIR systems Inc., Wilsonville, OR, USA) with a resolution of 640 × 512 pixels and a thermal sensitivity of
Sensors 2021, 21, 4505 3 of 12 Sensors 2021, 21, 4505 3 of 12 and weight of the subjects were 26.2 ± 1.32 years, 175 ± 0.89 cm, and 67.4 ± 2.79 kg, respec- respectively. The subjects tively. The subjects exposed exposed their brachium their brachium to air to air for 15 for min15before min before thermal thermal imageimage acqui- acquisition. sition. Figure1. Figure 1. Experimental Experimental setup for detection detection of of muscle muscle activation. activation.Thermal Thermalimages imagesofofthe thebrachium brachium were obtained using a thermal camera during the training sessions. were obtained using a thermal camera during the training sessions. 2.2. 2.2. Experimental Experimental Procedures Procedures The The subjects subjects performed performed training training movements movements to activate activate each each of of the the target target muscles. muscles. All All training was performed training was performedatatthe therate rateofof 1515 repetitions repetitions perper minute minute andandtimedtimed withwith a a met- metronome to ensure periodicity of movements. The total number of sets ronome to ensure periodicity of movements. The total number of sets per session, recov- per session, recovery ery time, time, and repetitions and repetitions per per set set varied were were varied betweenbetween experiments. experiments. Three Three different different exper- experiments were conducted to investigate the effects of isolated resistance training iments were conducted to investigate the effects of isolated resistance training on the skin- on the skin-surface surface temperature. temperature. A summary A summary of the of the three three training training conditions conditions assessedassessed in thisis in this study study is shown shown in Table 1. in Table 1. Table1.1.Summary Table Summaryofoftraining trainingconditions conditionsfor forthe theexperiments experiments(#: (#:Number). Number). Training TrainingWeight Weight Recovery Recovery ## of ofSets Sets per # of # of # of #Repetitions of Repetitions # of# of Target Target Experiment Experiment [kg] Time [s] per Session Sessions per Set Subjects Muscle [kg] Time [s] Session Sessions per Set Subjects Muscle I.I.Weight Weight 5 5and and1010 300 300 1 set 1 set 1 1 15 15 1 1 Biceps Biceps II. Duration 10 92 3 sets 1 8 1 Biceps II. Duration 10 92 3 sets 1 8 1 Biceps III. Muscle 5 or 10 92 3 sets 3 12 5 All Three III. Muscle 5 or 10 92 3 sets 3 12 5 All Three To To investigate investigate the the effects effectsofofweight weightvariation variation (experiment (experiment I) on I) on thethe subject’s subject’s skinskin sur- surface face temperature changes, the biceps was chosen as the target muscle. A single subject temperature changes, the biceps was chosen as the target muscle. A single subject performed performedone oneset setofoffull fullarm-curl arm-curlsessions, sessions,where whereeach eachset setincluded included 1515 repetitions (i.e., repetitions 6060 (i.e., s duration) to ensure that the training session fully stimulated the target muscle. s duration) to ensure that the training session fully stimulated the target muscle. This was This was followed by 5 min of recovery. Thermal videos of two sessions were acquired, while the followed by 5 min of recovery. Thermal videos of two sessions were acquired, while the training sessions were performed using 5 kg and 10 kg dumbbells separately. In addition, training sessions were performed using 5 kg and 10 kg dumbbells separately. In addition, the effects of consecutive sets of training (experiment II) on the subject’s skin-surface the effects of consecutive sets of training (experiment II) on the subject’s skin-surface tem- temperature were investigated for the biceps. The subject performed three sets of eight perature were investigated for the biceps. The subject performed three sets of eight repe- repetitions (i.e., 36 s duration including passive (no movement with weight) training for titions (i.e., 36 s duration including passive (no movement with weight) training for 4 s). 4 s). Each set was followed by 92 s of recovery. To fully stimulate the target muscle, a 10 kg Each set was followed by 92 s of recovery. To fully stimulate the target muscle, a 10 kg dumbbell was used. Thermal video of the entire session was recorded. Lastly, muscle dumbbell was used. Thermal video of the entire session was recorded. Lastly, muscle ac- activation from different target muscles (experiment III) was investigated. To observe tivation from different target muscles (experiment III) was investigated. To observe the the effects of the order of training on different target muscles, three sessions with three effects of the order of training on different target muscles, three sessions with three differ- different training orders, corresponding to activation of the three different target muscles, ent training orders, corresponding to activation of the three different target muscles, were were employed. The selected training orders were order #1: arm curl–lateral raise–kickback;
Sensors 2021, 21, 4505 4 of 12 Sensors 2021, 21, 4505 4 of 12 employed. The selected training orders were order #1: arm curl–lateral raise–kickback; order #2: lateral raise–kickback–arm curl; and order #3: kickback–arm curl–lateral raise; hence,#2:the order corresponding lateral muscle activations raise–kickback–arm were#3: curl; and order biceps–deltoid–triceps; kickback–arm curl–lateral deltoid–tri- raise; ceps–biceps; and triceps–biceps–deltoid. Each subject performed three sets hence, the corresponding muscle activations were biceps–deltoid–triceps; deltoid–triceps– of 12 repeti- tions (i.e., biceps; and48triceps–biceps–deltoid. s duration), followed byEach92 s subject of recovery for eachthree performed session. setsEach of 12subject chose repetitions either (i.e., 48 the 5 kg or 10 s duration), kg dumbbell followed by 92 saccording of recoveryto for their eachability to withstand session. the chose Each subject full training either session. the 5 kg orEach training 10 kg session dumbbell was repeated according to theirfor bothto ability the left and right withstand arms, the full and asession. training total 70 sessions Each weresession training selected. was repeated for both the left and right arms, and a total 70 sessions were selected. 2.3. Data Processing 2.3. Data Processing All images were aligned so that the biceps were located on the left sides of the images All images (Figure 2a). Then,werethealigned brachium so that areathe wasbiceps were located manually croppedon the left2b). (Figure sides of the After imagesa applying (Figure 2a). Then, the brachium area was manually cropped (Figure moving average filter, Otsu’s method [20] for global thresholding was applied to segment2b). After applying a moving average filter, Otsu’s method [20] for global thresholding was the brachium areas (Figure 2c). Further segmentation was manually performed, if neces- applied to segment the brachium sary. areasof(Figure The region interest 2c). Further (ROI) wassegmentation was manually then evenly divided into 20performed, horizontalifand necessary. 60 ver- The ticalregion of (Figure sections interest2d).(ROI) was map A heat thenwasevenly dividedbyinto generated 20 horizontal calculating the meanandtemperature 60 vertical sections from each (Figure divided2d).region A heatinmap the was ROIgenerated (Figure 2e)byandcalculating displayed theusing meana temperature color map (Figurefrom each 2f). divided region in the ROI (Figure 2e) and displayed using a color map (Figure 2f). To Todetect detect muscle muscle activation, activation, the the thermal thermalimages imagesatatthe thebeginning beginningand and endend of of eacheachset set were chosen from each training set. Thus, a total of six images were chosen from each training set. Thus, a total of six images were used from the three were used from the three sets insets in a single a single session.session. Using Using the thermal the thermal images,images, heatwere heat maps maps were generated generated and denoted and denoted as (S1), as (S1), (S2), and(S2), and (E1), (S3) and (S3) and (E2),(E1), and (E2), andthe (E3) for (E3) for the beginning beginning and end ofand eachendset,ofrespec- each set, respectively. Thus, a total of 420 heat maps were generated tively. Thus, a total of 420 heat maps were generated out of the 70 sessions. out of the 70 sessions. Figure Figure2. Visualization of 2. Visualization ofthe theprocess processofofanalysis analysisofof a single a single image: image: (a) original (a) original thermal thermal image, image, (b) (b) cropped image around the brachium, (c) image after segmentation, (d) 20 × 60 divided cropped image around the brachium, (c) image after segmentation, (d) 20 × 60 divided regions, (e) regions, heat (e) map heat mapwith three with selected three regions selected for for regions the the target muscles, target and and muscles, (f) colorized heat heat (f) colorized map map of theoftar- the get muscles. target muscles. To Toanalyze analyzethe thetemperature temperaturechangeschangesfromfromthe thetarget targetmuscle muscleregions, regions,each eachheat heatmap map was was divided divided into three three regions, regions,as asshown shownininFigure Figure 2e.2e. Based Based on positions on the the positions ofmus- of the the muscles, it was cles, it was assumed assumed thatthat thethe upper-halfregion upper-half regionwas wasthe the deltoid, deltoid, the lower-left lower-leftquadrant quadrant was wasthethebiceps, biceps,and andthe thelower-right lower-rightquadrant quadrantwas wasthe thetriceps. triceps. The Themean meanvalue valueofofeach each region regionrepresents representsthe thetemperature temperatureof ofthe thecorresponding correspondingmuscle. muscle.Statistical Statisticalanalyses analyseswere were performed performedusingusingthe theKruskal–Wallis Kruskal–Wallistest testto toexamine examinethe thetemperature temperaturedifference differencefor foreach each target muscle between the three training target muscle between the three training orders. orders.
Sensors 2021, 21, 4505 5 of 12 2.4. Classification of Target Muscle Activation The ResNet-18 [21] network was used to classify target muscle activation using the difference heat maps (i.e., temperature differences) from the proposed method. To train the network, a total of 134 difference heat maps were generated from the differences between the heat map pairs, which are (S1-E3) and (S1-S3). The target classification was divided into four classes, namely biceps, triceps, deltoid, and vague. The “vague” class implies that no valid muscle activation was observed. The test set was 10% of randomly selected data, thus resulting in 121 training data and 13 test data. The summary of the dataset is shown in Table 2. For augmentation of the training data, the heat maps were processed by rotation from −1◦ to 1◦ in steps of 1◦ , Gaussian noise addition with a mean of 0.07, scaling by 0.9 and 1.1 times, and resizing to a resolution of 244 × 244 pixels, resulting in 2178 training images. The network model was initialized using the pretrained weights with the ImageNet dataset [22] for ResNet-18 provided in MATLAB (Mathworks Inc., Natick, MA, USA). Then, the model was trained by finetuning with a learning rate of 1 × 10−4 and batch size of 16 for 50 epochs. The Adam optimizer and cross-entropy loss were used for training. Table 2. Summary of the dataset used for training and test for classification. Dataset Biceps Triceps Deltoid Vague Total Train 33 19 31 38 121 Test 4 3 2 4 13 3. Results 3.1. Effects of Weight, Duration, and Muscle Activation for Training Figure 3 shows the changes in skin surface temperatures from the biceps with vary- ing weights (experiment I (Figure 3a) and in consecutive sets of training (experiment II (Figure 3b). For the measurement with a 5 kg dumbbell (dotted line in Figure 3a), the initial biceps skin temperature was measured as 36.07 ◦ C. After one minute of training, the biceps skin temperature was still 36.07 ◦ C. After five minutes of recovery, the skin temperature reached a maximum of 36.34 ◦ C. While the average rate of increase of the skin surface temperature on the biceps during training was negligible, the skin surface temperature during recovery was 0.05 ◦ C/min. With a 10 kg dumbbell (solid line in Figure 3a), the skin surface temperature was 35.98 ◦ C initially and 36.11 ◦ C at the end of training. After five minutes of recovery, the temperature reached up to 37.24 ◦ C. The average elevation rate was 0.13 ◦ C/min during exercise and 0.47 ◦ C/min during 3 min of recovery. It was there- fore observed that the increase in the skin surface temperature occurred during recovery. Moreover, the skin temperature reached a plateau and decreased after enough recovery time (3 min in this study). The difference in the average elevation rate during 3 min of recovery with the 5 kg and 10 kg weights was 0.46 ◦ C/min. Figure 3b shows skin surface temperature changes during three consecutive sets of training with recovery times between each set. The temperature gradually rose during rest, which matches with the results from a single set of training in Figure 3a. Further, it was observed that the skin surface heated up as more sets of training were performed. The overall average elevation rate was 0.52 ◦ C/min. In each recovery period, the average elevation rates of the skin surface temperatures were 0.51, 0.42, and 0.59 ◦ C/min, as shown in Figure 3b. The mean skin surface temperatures during exercise were 32.90, 34.00, and 35.07 ◦ C for each training set. The mean skin surface temperatures during recovery were 33.42, 34.65, and 35.64 ◦ C after each training set.
Sensors 2021, 21, 4505 6 of 12 Sensors 2021, 21, 4505 6 of 12 Sensors 2021, 21, 4505 6 of 12 (a) Exp. I. Weight Variation (b) Exp. II. Consecutive Set 38 (a) Exp. I. Weight Variation 36.5 (b) Exp. II. Consecutive Set 38 10kg 36.5 0.59℃/min 10kg 5kg 36 0.59℃/min 5kg 36 37.5 35.5 37.5 35.5 0.42℃/min 0.47℃/min 35 0.42℃/min 0.47℃/min 35 37 37 34.5 34.5 34 0.51℃/min 36.5 34 0.51℃/min 36.5 0.05℃/min 0.05℃/min 33.5 33.5 33 36 33 36 A R 32.5 A R A R A R A R 32.5 A R A R A R 0 60 120 180 240 300 360 0 36 128 164 256 292 360 0 60 120 180 240 300 360 0 36 128 164 256 292 360 Time Time (s) (s) Time(s) Time (s) Figure Figure3.3.Change Changein inskin skinsurface temperature surfacetemperature from temperaturefrom biceps bicepsinin frombiceps in (a) aa session session of aa single single set with with 5kg kgand and1010kg kgdumbbells dumbbells Change in skin surface (a)(a) a session of of a single set set with 5 kg5 and 10 kg dumbbells and and and in (b) a session of three consecutive sets with a 10 kg dumbbell. (“A” and “R” denote active training and recovery in (b)ina (b) a session session of three of three consecutive consecutive setsawith sets with 10 kga dumbbell. 10 kg dumbbell. (“A” and(“A” “R”and “R” active denote denotetraining active training and recovery and recovery periods, periods, periods,respectively.). respectively.). respectively.). Figure Figure44 shows Figure shows examples examples of the heat of the heat maps maps for for the the three threedifferent differentmuscle muscleactivations activations (i.e., (a) (i.e., (a) biceps, (a)biceps, (b) biceps,(b) triceps, (b)triceps, and triceps,and and(c) deltoid) (c)(c) deltoid) from from deltoid) experiment experiment from experiment III. Column III. Column III. Column (i) shows (i) shows theheat the (i) shows heat the maps maps of heat mapsof the beginning theofbeginning the beginningof of each of training each training set (S1-S3), set each training (S1-S3), column column set (S1–S3), (ii)shows (ii) column shows theheat the (ii) shows heat maps themaps heat ofofthe mapsthe end end of of each each training training set set (E1-E3) (E1-E3) from from a a single single session, session, and and column column (iii) (iii) of the end of each training set (E1–E3) from a single session, and column (iii) shows the shows shows the the difference difference heat heat maps mapsfrom difference the the (S1-E2), frommaps heat (S1-E2), from the(S1-S3), (S1-S3), (S1–E2), and and (S1-E3)and (S1-E3) (S1–S3), pairs. pairs. Thedifference The (S1–E3) difference pairs. Theheat heatmaps mapsare difference areshown heat shown maps in in Figure areFigure 4a–c shown4a–c (iii). (iii). 4a–c (iii). in Figure Figure Figure4.4. Figure 4.Heat Heatmaps Heat mapsfrom maps frommuscle from muscleactivation muscle activationofof activation of(a) (a)biceps, (a) biceps, biceps, (b) (b) triceps, triceps, (b) and triceps, and (c)(c) and deltoid. deltoid. (c) Column Column deltoid. Column(i): (i): (i):heat heat maps maps heat ofofthe of the maps thebegin- beginning begin- ning ning of each of each of each training training training set (S1-S3), set (S1-S3), set (S1–S3), column column column (ii): heat (ii):maps (ii): heat maps heat maps of the of the of the end end of end of eachof each training each training training set (E1-E3) set (E1-E3) set (E1–E3) from a single from session, from a single session, a single and session, and and column column column (iii): difference (iii): difference heat maps heatfrom maps from (S1-E3), from (S1-E3), (S1-S2), and (S1-S3) pairs. (iii): difference heat maps (S1–E3), (S1–S2),(S1-S2), and (S1-S3) and (S1–S3) pairs. pairs. Difference Difference heat Difference heat maps heat maps were generated maps were generated from generated from various from various heat heatmap variousheat mappairs, map pairs,including pairs, including(S1-E3), including (S1–E3), (S1-E3), (S1-S3), (S1-E2), (S1-S2), and (S1-E1). Table 3 shows the numbers of pairs that producethe (S1-S3), (S1–S3), (S1-E2), (S1–E2), (S1-S2), (S1–S2),and and (S1-E1). (S1–E1).Table 3 Table shows 3 the shows numbers the numbersof pairs of that pairs produce that producethe highest highest temperature the highest temperature differences temperature differences among differences among the the various among heat the various various map heatheat pairs mapmap from pairs pairs from each from session. eacheach The session. session. The average difference The average for S1-E3 difference of the entire for S1-E3 collected data data was 0.86 °C, ◦ C,the and highest tem- average difference for S1-E3 of theofentire the entire collected collected data was was 0.86 0.86 °C, and and the highest the highest tem- perature temperaturedifference was difference 0.97 °C. ◦ The higher the temperature difference, the clearer the perature difference waswas0.970.97 °C. C. The The higherthe higher thetemperature temperature difference, difference, thethe clearer clearerthe the
Sensors 2021, 21, 4505 7 of 12 Sensors 2021, 21, 4505 7 of 12 differentiation of the selected muscles. It was observed that 85% of the highest tempera- ture difference was acquired from the heat map pairs of the first and third sets of each session. differentiation of the selected muscles. It was observed that 85% of the highest temperature difference was acquired from the heat map pairs of the first and third sets of each session. Table 3. Counts of different periods showing the highest temperature differences from heat map pairs. Table 3. Counts of different periods showing the highest temperature differences from heat map pairs. Heat Map Pairs Count Percentage Heat Map Pairs Count Percentage (S1-E3) 36 51% (S1-E3) (S1-S3) 36 24 51% 34% (S1-S3) 24 34% (S1-E2) 4 5% (S1-E2) 4 5% (S1-S2) (S1-S2) 22 3% 3% (S1-E1) (S1-E1) 44 5% 5% Figure 55shows Figure showsvarious variousheat heatmapmapexamples examplesgenerated generatedfrom fromthetheexperiments experimentsin inthis this study. Figure 5a is the control case where the subjects performed training with study. Figure 5a is the control case where the subjects performed training with no weights no weights (i.e., passive (i.e., passive training); training); nono local local heating heating was wasobserved observedininthis thiscase. case. Figure Figure 5b–d 5b–d clearly clearly demonstrates local heating of the target muscles with proper training demonstrates local heating of the target muscles with proper training for each targetfor each target mus- cle activation. muscle Figure activation. 5f shows Figure an example 5f shows an exampleof unexpected muscle of unexpected activation muscle from from activation over- weighted training, where the subject swung the elbow excessively while over-weighted training, where the subject swung the elbow excessively while targeting targeting the bi- ceps. Additionally, in the case of Figure 5e, the subject performed an inaccurate the biceps. Additionally, in the case of Figure 5e, the subject performed an inaccurate kick-back session. The kick-back resulting session. images show The resulting imagesthat the temperature show increased that the temperature throughout increased the bra- throughout chium, especially at the posterior deltoid (i.e., upper-right area). the brachium, especially at the posterior deltoid (i.e., upper-right area). Figure5.5.Heat Figure Heatmap mapexamples examplesfrom from(a) (a)control controlfrom frompassive passivetraining, (b–d) training, proper (b–d) training proper forfor training biceps, bi- triceps, and deltoids, respectively, (e) over-weighted training, and (f) incorrect posture of training. ceps, triceps, and deltoids, respectively, (e) over-weighted training, and (f) incorrect posture of training. 3.2. Statistical Analysis Figure 6 shows 3.2. Statistical Analysisthe overall average of temperature differences between the regions of target and nontarget muscles during the exercise session. Overall, the heat map pair Figure 6 shows the overall average of temperature differences between the regions (S1)-(E3) showed the most significant temperature difference, as analyzed from Table 2. of target and nontarget muscles during the exercise session. Overall, the heat map pair The results in Figure 6 match well with those in Figure 3, where the surface temperature (S1)-(E3) showed the most significant temperature difference, as analyzed from Table 2. of the target muscle gradually increased during a session with three training and rest The results in Figure 6 match well with those in Figure 3, where the surface temperature periods. In the case of the biceps (Figure 6a), the temperature difference was higher when of the target muscle gradually increased during a session with three training and rest pe- compared with the deltoid (dotted line) than with the triceps (solid line). For the triceps riods. case In the6b), (Figure case ofoverall the the biceps (Figure 6a), temperature the temperature difference was lower difference compared to wasthehigher when other cases. compared with the deltoid (dotted line) than with the triceps (solid line). For the triceps For the deltoid case (Figure 6c), the trend with the triceps (solid line) was similar to that of casebiceps the (Figure 6b), line). (dotted the overall temperature difference was lower compared to the other cases. For the deltoid case (Figure 6c), the trend with the triceps (solid line) was similar to that of the biceps (dotted line).
Sensors Sensors 2021, Sensors2021, 21, 2021,21, 4505 21,4505 4505 888of of 12 of 12 (a) (a)Biceps Biceps (b) (b)Triceps Triceps (c) (c)Deltoid Deltoid Bi Bi--Tri Tri Tri Tri--Bi Bi Delt Delt--Bi Bi 0.8 0.8 Bi Bi--Delt Delt 0.8 0.8 Tri Tri--Delt Delt 0.8 0.8 Delt Delt--Tri Tri 0.6 0.6 0.6 0.6 0.6 0.6 0.4 0.4 0.4 0.4 0.4 0.4 0.2 0.2 0.2 0.2 0.2 0.2 00 00 00 Figure Figure6. Figure 6.The 6. Theaverage The averagetemperature average temperaturedifference temperature differencebetween difference betweentarget between targetmuscles target musclesand muscles andnon-target and non-targetmuscles non-target muscleswhen muscles whentargeting when targeting(a) targeting (a)biceps, (a) biceps, biceps, (b) (b)triceps, triceps,and and(c) (c)deltoid. deltoid.Error Errorbars barsdenote denotestandard standarddeviation. deviation. (b) triceps, and (c) deltoid. Error bars denote standard deviation. Figure Figure 77 shows shows the the median median and and variance variance values values ofof the the temperature temperature differences differences forfor each target each target muscle targetmuscle from musclefrom thethe from the three threethree different different different training training training orders. orders.orders. The The Kruskal–Wallis Kruskal–Wallis The Kruskal–Wallis test test showedtest showed showed that that no no significant that no significant significant differencesdifferences (p-value =(p-value differences (p-value == 0.57, 0.57, 0.62, 0.62, 0.57,0.40 and forand 0.62, and 0.40 for for biceps, 0.40triceps, biceps, and triceps, biceps, triceps, deltoid, and and deltoid, deltoid, respectively) respectively) respectively) were were found were found among found the among among three the three three different the training different orders. training different training orders. orders. The results The The re- demonstrate re- sults sults demonstrate demonstrate that the that thatthe order of muscle order orderof theactivationofmuscle muscle activation activationhad had insignificant had insignificant insignificant effects effects effectson on the temperature onthe thetem- tem- changes perature perature changes changesduring during exercise. duringexercise. exercise. Temperature Difference Temperature [℃] Difference [℃] Figure 7. Temperature differences for target muscles (a) biceps, (b) triceps, and (c) deltoid from three different training Figure 7.7. Temperature Figure (order Temperature differences differences for for target target muscles (a) (a) biceps, biceps, (b) (b) triceps, and (c) (c) deltoid from from three three different different training orders #1: biceps–deltoid–triceps, ordermuscles triceps, #2: deltoid–triceps–biceps, andandorder deltoid #3: triceps–biceps–deltoid). training orders orders(order (order#1: #1:biceps–deltoid–triceps, biceps–deltoid–triceps,order order#2: #2:deltoid–triceps–biceps, deltoid–triceps–biceps,and andorder order#3: #3:triceps–biceps–deltoid). triceps–biceps–deltoid). 3.3. ClassificationofofMuscle 3.3. Muscle Activation 3.3.Classification Classification of MuscleActivation Activation The The training accuracy and loss function of the network training are shown ininFigure 8. Thetraining trainingaccuracy accuracyandandloss lossfunction functionof ofthe thenetwork networktraining trainingare areshown shown inFigure Figure After 50 epochs, 100% training accuracy and 92.3% test accuracy were achieved. Figure 9 8. 8.After After5050epochs, epochs,100% 100%training trainingaccuracy accuracyandand92.3% 92.3%test testaccuracy accuracywerewereachieved. achieved.Figure Figure shows the prediction probabilities for the classification of difference heat maps for various 99shows showsthe theprediction predictionprobabilities probabilitiesforforthe theclassification classificationofofdifference differenceheat heatmaps mapsforforvari- vari- muscle activations. Each target muscle activation showed the highest prediction probability ous ousmuscle muscleactivations. activations.Each Eachtarget targetmuscle muscleactivation activationshowed showedthe thehighest highestprediction predictionprob- prob- from the correct class (biceps: 99.99% (Figure 9a), deltoid: 98.09% (Figure 9b), triceps: ability ability from from the the correct correct class class (biceps: (biceps: 99.99% 99.99% (Figure (Figure 9a), 9a), deltoid: deltoid: 98.09% 98.09% (Figure (Figure 9b), 9b), tri- tri- 99.98% (Figure 9c), vague: 99.84% (Figure 9d). Although insufficient training and test ceps: ceps: 99.98% 99.98% (Figure (Figure 9c), 9c), vague: vague: 99.84% 99.84% (Figure (Figure 9d). 9d). Although Although insufficient insufficient training training and and datasets were utilized, the results clearly demonstrate that classification of muscle activation test test datasets datasets were were utilized, utilized, the the results results clearly clearly demonstrate demonstrate that that classification classification of of muscle muscle can be easily achieved from the trained network using the difference heat maps. activation activationcan canbe beeasily easilyachieved achievedfromfromthe thetrained trainednetwork networkusing usingthe thedifference differenceheat heatmaps. maps.
Sensors Sensors 2021,21, Sensors2021, 2021, 21,4505 21, 4505 4505 999of of 12 of12 12 Figure8.8.Summary Figure Summaryofofnetwork networktraining: training:(a) (a)training trainingaccuracy accuracyand and(b) (b)loss lossfunction. function. Figure 9. Classification of muscle activation. (a–d): test heat maps of three different muscle activations Figure9.9.Classification Figure Classificationofofmuscle muscleactivation. activation.(a–d): (a–d):test testheat heatmaps mapsofofthree threedifferent differentmuscle muscleactiva- activa- and vague tions case with andvague vague casepredicted probabilities withpredicted predicted from the probabilities fromnetwork, respectively. thenetwork, network, respectively. tions and case with probabilities from the respectively. 4. Discussion 4.4.Discussion Discussion This study demonstrated the correlation between isolated resistance training of three separate Thismuscles This studydemonstrated study demonstrated of the brachium thecorrelation the correlation and between between temperatures isolated isolated of resistancetraining resistance their corresponding training ofofthree locations three on separate separate muscles muscles of of the the brachium brachium and and temperatures temperatures ofof the skin surface. Three muscles of the brachium (i.e., biceps, triceps, and deltoid) were their their corresponding corresponding locations locations on on the the skin skin surface. surface. Three Three muscles muscles of of the the brachium brachium (i.e., (i.e., selected owing to their easy isolation for resistance training movements and the anatomy biceps, biceps, triceps, triceps, and and deltoid) deltoid) were were selected selected of owingdistinguishable owing the muscles tototheir theireasy easyisolation isolation underfor for resistance resistance thermal training training imaging. movements Themovements main findings andof and the the anatomy anatomy this study of of the arethe muscles as muscles distinguishable follows: distinguishable (1) weight variation under underduring thermal thermaltraining imaging. imaging. The The main affects main findings findings the change in of of this study thissurface skin study areas are asfollows: follows: temperature; (1) (1) (2) weightvariation weight consecutive variation training during during trainingheating training sets increase affectsthe affects thethe of change change skin in inskin skinsurface surface; surface and (3)tem- tem- the perature; perature; order (2)consecutive (2) of trainingconsecutive trainingsets training has an insignificant sets increase increase effect heatingof heating on isolating ofthe theskin muscle skin surface;and surface; activation. and(3) (3)thetheor-or- derofof der training Ittraining was hasan has observed aninsignificant insignificant that temperature effectdifferences effect onisolating on isolating muscle inmuscle activation. activation. the thermal images were highest mostly ItItwas wasobserved betweenobserved the start that thatand temperature end of training temperature differences sessions differences inin(Table thethermal the thermal images 2). Thisimages does not were were match highest well highest mostly with mostly between results between from the the startand previous start and endofoftraining studies, end training which sessions(Table considered sessions (Table training 2).This 2). Thisdoes with does highnot not matchwell intensity match well and withresults duration with results fromprevious leading from previous to studies, perspiration studies, andwhich which muscleconsidered training fatiguetraining considered [23–25].withwith highstudy, In high this intensity intensity theandand effectsdu- du- ration from ration leadingtotoperspiration perspiration leading perspiration and muscle and andmuscle fatiguemuscle fatigue werefatigue minimized [23–25]. [23–25].using Inthis In thisstudy, recoverystudy, time thebetween the effectsfrom effects from the training perspiration perspiration sets.and Overall, and muscle muscle there werewere fatigue fatigue relatively were minimized minimized subtleusing temperature using recovery recovery changes timebetween time duringthe between training the train- train- compared ing sets. to that Overall, during there recovery were time, relatively as shown subtle in Figures temperature ing sets. Overall, there were relatively subtle temperature changes during training com- 3 and changes 4. This during can be explained training com- by paredthe process paredtotothat thatduringof human duringrecovery metabolism. recoverytime, time,as asshown When muscles shownininFiguresFigures33andproduce and4.4.This energy Thiscan canbe by movement, beexplained explainedby by PCr the hydrolysis theprocess process is initially ofofhuman human dominant metabolism. metabolism. (~10 muscles When When s), followed muscles by glycolysis produce produce energyby energy (~50 by s). Subsequently, movement, movement, PCrhy- PCr hy- oxidation drolysisisiswill drolysis dominate initially initially dominantenergy dominant (~10 production (~10 s),followed s), followed [9].byAsglycolysis by heat production glycolysis (~50s). (~50 s).occurs mainly during Subsequently, Subsequently, oxida- oxida- the tionoxidation tion willdominate will process, dominate exercise energy energy durations[9]. production production longer [9]. As than Asheat 60 s are expected heatproduction production occurs occurs tomainly contribute mainly to heat during during the the production. oxidation In this process, study, exercise the training durations time longer for each than 60 set s oxidation process, exercise durations longer than 60 s are expected to contribute to heat was are up expectedto 60 tos, so the contribute oxidation to heat process production. production. wasIn not In thisconsidered. this study,the study, As thetraining it is time training known timefor forthat each each only 2% wasof setwas set upheat up toto60 produced 60 s,s,sosothe from theoxidation oxidation the muscle processwas process is instantly wasnot dissipated notconsidered. considered.As through Asititisisknown the knownthat skin while thatonly only2%the rest 2%ofofheat is accumulated heatproduced producedfrom in fromthe the muscle themuscle muscle
Sensors 2021, 21, 4505 10 of 12 and slowly perfused through skin, inducing vasodilation near the activated muscle [10], this heat transfer causes delays in temperature change on the skin surface. Thus, it is expected that the temperature rise during the recovery time would be more noticeable than that during training, as observed in this study. Overall, the results in Figure 6 show that the biceps region exhibited the most signifi- cant rise in temperature. This can be explained by the presence of major veins (i.e., cephalic vein) in the biceps region. The heat accumulated during training was dissipated through the veins. In contrast, in other regions with no major veins, heat dissipation was relatively low (Figure 6b,c). Owing to the thickest skinfold with no major veins, the heat dissipation from the triceps region was the lowest among the three muscles [26], which was also ob- served in this study (Figure 6b). This is presumably because the deltoid muscles function as synergist muscles during motion. Although the overall temperature increased, there were several temperature decreases for the deltoid (Figure 6c). During the lateral raise motion to activate the deltoid muscle, the subjects swung their arms away from the body until their arms became parallel with the floor. This movement may cause cooling via respiration. As shown in Figure 5a–f, accurate training postures result in localized heating of the skin directly above the target muscles. Thus, accurate training postures for isolated resistance training can be clearly visualized using thermal imaging. Utilizing the difference heat maps generated in this study, a convolutional neural network (ResNet-18 in this study) was trained and tested for classification of muscle activation. Although there were not sufficient data to discuss the accuracy or performance of the network, the test results of the network clearly demonstrate the feasibility of the proposed approach to differentiate muscle activation. The proposed technique can be further utilized in physical therapy, which should be performed with proper muscle activation. In this study, the number of subjects was limited, and the number of variations was small. There are various conditions of the subject, such as body mass composition, muscle mass, training skill, color and thickness of skin, thickness of skinfold, and age, that may affect the result of this study. Thus, further investigations by performing experiments on various subject groups are necessary [24,27]. In order to achieve repeatability of the measurements, it is necessary to minimize changes in distance and angle between the camera and the subject. Thus, motion tracking of the target area could be implemented. Although the proposed approach is a noninvasive and noncontact method, the surface of the skin needs to be exposed during the measurement. As only two absolute loads (5 and 10 kg) were utilized, which can induce different relative intensities of work depending on the subject’s muscle strength, it is also necessary to consider the absolute intensity of work correlated to muscle activation. Further investigation on validating the accuracy and efficacy of the isolated muscle training sessions with the proposed approach will be our future work. 5. Conclusions The activation of three different brachium muscles during resistance training was detected using infrared thermal imaging and analyzed using heat maps. The proposed approach in this study demonstrated that the differences in heat maps could be used to effectively differentiate the target muscles. Further investigations are necessary to validate this approach on subject groups with various conditions and to expand it to other target muscles. Future studies will also seek to improve the accuracy and efficacy of the proposed approach for muscle isolation. Author Contributions: Conceptualization, H.J., J.S., K.S., and S.P.; methodology, H.J., J.S., and K.S.; software, H.J., J.S., K.S., and D.K.; validation, S.P.; resources, S.P. and D.K.; writing—original draft preparation, H.J., J.S., and K.S.; writing—review and editing, S.P.; supervision, S.P. All authors have read and agreed to the published version of the manuscript.
Sensors 2021, 21, 4505 11 of 12 Funding: This research was supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science and ICT under grant number NRF-2020R1A2C1011889. Institutional Review Board Statement: Not applicable. Informed Consent Statement: Not applicable. Data Availability Statement: The data presented in this study are available upon request from the corresponding author. Conflicts of Interest: The authors declare no conflict of interest. References 1. Scott, B.R.; Duthie, G.M.; Thornton, H.R.; Dascombe, B.J. Training Monitoring for Resistance Exercise: Theory and Applications. Sports Med. 2016, 46, 687–698. [CrossRef] [PubMed] 2. Marston, K.J.; Peiffer, J.J.; Newton, M.J.; Scott, B.R. A comparison of traditional and novel metrics to quantify resistance training. Sci. Rep. 2017, 7, 5606. [CrossRef] [PubMed] 3. McBride, J.M.; McCaulley, G.O.; Cormie, P.; Nuzzo, J.L.; Cavill, M.J.; Triplett, N.T. Comparison of Methods to Quantify Volume during Resistance Exercise. J. Strength Cond. Res. 2009, 23, 106–110. [CrossRef] [PubMed] 4. Haff, G.G. Quantifying Workloads in Resistance Training: A Brief Review; UK Strength and Conditioning Association: Lytham, UK, 2010; p. 19. 5. Chiu, L.Z.F. Biomechanical Methods to Quantify Muscle Effort during Resistance Exercise. J. Strength Cond. Res. 2018, 32, 502–513. [CrossRef] 6. Taha, Z.; Lee, C.M.; Ahamed, N.U.; Joseph, S.; Omar, S.F.S. Performance Analysis in Strength Training: An Innovative Instrumen- tation. Procedia Eng. 2016, 147, 455–460. [CrossRef] 7. Gielen, F.L.H.; Wallinga-de Jonge, W.; Boon, K.L. Electrical conductivity of skeletal muscle tissue: Experimental results from different musclesin vivo. Med. Biol. Eng. Comput. 1984, 22, 569–577. [CrossRef] [PubMed] 8. Kamen, G.; Gabriel, D.A. Essentials of Electromyography; Human Kinetics: Leeds, UK, 2010. 9. Krustrup, P.; Ferguson, R.A.; Kjaer, M.; Bangsbo, J. ATP and heat production in human skeletal muscle during dynamic exercise: Higher efficiency of anaerobic than aerobic ATP resynthesis. J. Physiol. 2003, 549 Pt 1, 255–269. [CrossRef] 10. González-Alonso, J.; Quistorff, B.; Krustrup, P.; Bangsbo, J.; Saltin, B. Heat production in human skeletal muscle at the onset of intense dynamic exercise. J. Physiol. 2000, 524, 603–615. [CrossRef] 11. Ali, S.S.; Khan, A.Y.; Michael, S.G.; Tankha, P.; Tokuno, H. Use of Digital Infrared Thermal Imaging in the Electromyography Clinic: A Case Series. Cureus 2019. [CrossRef] [PubMed] 12. Chudecka, M.; Lubkowska, A. Temperature changes of selected body’s surfaces of handball players in the course of training estimated by thermovision, and the study of the impact of physiological and morphological factors on the skin temperature. J. Therm. Biol. 2010, 35, 379–385. [CrossRef] 13. Priego Quesada, J.I.; Carpes, F.P.; Bini, R.R.; Salvador Palmer, R.; Pérez-Soriano, P.; Cibrián Ortiz de Anda, R.M. Relationship between skin temperature and muscle activation during incremental cycle exercise. J. Therm. Biol. 2015, 48, 28–35. [CrossRef] [PubMed] 14. Escamilla-Galindo, V.L.; Estal-Martínez, A.; Adamczyk, J.G.; Brito, C.J.; Arnaiz-Lastras, J.; Sillero-Quintana, M. Skin temperature response to unilateral training measured with infrared thermography. J. Exerc. Rehabil. 2017, 13, 526–534. [CrossRef] [PubMed] 15. Zagrodny, B.; Wojnicz, W.; Ludwicki, M.; Awrejcewicz, J. Could Thermal Imaging Supplement Surface Electromyography Measurements for Skeletal Muscles? IEEE Trans. Instrum. Meas. 2021, 70, 1–10. [CrossRef] 16. Daud, S.A.; Elamvazuthi, I.; Bt Zulkifli, Z.; Parasuraman, S.; Ahamed Khan, M.K.A. Analysis of thermal imaging in determining muscle contractions of upper extremities. In Proceedings of the 2017 IEEE 3rd International Symposium in Robotics and Manufacturing Automation (ROMA), Kuala Lumpur, Malaysia, 19–21 September 2017; IEEE: Piscataway, NJ, USA, 2017; pp. 1–5. 17. Formenti, D.; Ludwig, N.; Rossi, A.; Trecroci, A.; Alberti, G.; Gargano, M.; Merla, A.; Ammer, K.; Caumo, A. Skin temperature evaluation by infrared thermography: Comparison of two image analysis methods during the nonsteady state induced by physical exercise. Infrared Phys. Technol. 2017, 81, 32–40. [CrossRef] 18. Ferreira, J.J.A.; Mendonça, L.C.S.; Nunes, L.A.O.; Andrade Filho, A.C.C.; Rebelatto, J.R.; Salvini, T.F. Exercise-Associated Thermographic Changes in Young and Elderly Subjects. Ann. Biomed. Eng. 2008, 36, 1420–1427. [CrossRef] [PubMed] 19. Kuniszyk-Joóźkowiak, W.; Jaszczuk, J.; Czaplicki, A. Changes in electromyographic signals and skin temperature during standardised effort in volleyball players. Acta Bioeng. Biomech. 2018. [CrossRef] 20. Otsu, N. A threshold selection method from gray-level histograms. IEEE Trans. Syst. Man Cybern. 1979, 9, 62–66. [CrossRef] 21. He, K.; Zhang, X.; Ren, S.; Sun, J. Deep Residual Learning for Image Recognition. arXiv 2015, arXiv:1512.03385. 22. Deng, J.; Dong, W.; Socher, R.; Li, L.-J.; Kai, L.; Li, F.-F. ImageNet: A large-scale hierarchical image database. In Proceedings of the 2009 IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops (CVPR Workshops), Miami, FL, USA, 20–25 June 2009; IEEE: Piscataway, NJ, USA, 2009; pp. 248–255.
Sensors 2021, 21, 4505 12 of 12 23. Hadžić, V.; Širok, B.; Malneršič, A.; Čoh, M. Can infrared thermography be used to monitor fatigue during exercise? A case study. J. Sport Health Sci. 2019, 8, 89–92. [CrossRef] [PubMed] 24. Fernandes, A.d.A.; Amorim, P.R.d.S.; Brito, C.J.; de Moura, A.G.; Moreira, D.G.; Costa, C.M.A.; Sillero-Quintana, M.; Marins, J.C.B. Measuring skin temperature before, during and after exercise: A comparison of thermocouples and infrared thermography. Physiol. Meas. 2014, 35, 189–203. [CrossRef] [PubMed] 25. Formenti, D.; Ludwig, N.; Gargano, M.; Gondola, M.; Dellerma, N.; Caumo, A.; Alberti, G. Thermal Imaging of Exercise- Associated Skin Temperature Changes in Trained and Untrained Female Subjects. Ann. Biomed. Eng. 2013, 41, 863–871. [CrossRef] [PubMed] 26. Eaton-Evans, J. Nutritional Assessment: Anthropometry. In Encyclopedia of Human Nutrition; Elsevier: Amsterdam, The Nether- lands, 2013; pp. 227–232. 27. Costello, J.T.; Culligan, K.; Selfe, J.; Donnelly, A.E. Muscle, Skin and Core Temperature after −110 ◦ C Cold Air and 8 ◦ C Water Treatment. PLoS ONE 2012, 7, e48190. [CrossRef] [PubMed]
You can also read