Assessment of Hip and Knee Joints and Implants Using Acoustic Emission Monitoring: A Scoping Review

Objectives: Population ageing and the subsequent increase of joint disorders prevalence requires the development of non-invasive and early diagnostic methods to enable timely medical assistance and promote healthy aging. Over the last decades, acoustic emission (AE) monitoring, a technique widely used in non-destructive testing, has also been introduced in orthopedics as a diagnostic tool. This review aims to synthesize the literature on the use of AE monitoring for the assessment of hip and knee joints or implants, highlighting the practical aspects and implementation considerations. Methods: this review was conducted as per the PRISMA statement for scoping reviews. All types of studies, with no limits on date of publication, were considered. Articles were assessed and study design parameters and technical characteristics were extracted from relevant studies. Results: conducted search identified 1379 articles and 64 were kept for charting. Seven additional articles were added at a later stage. Reviewed works were grouped into studies on joint condition assessment, implant assessment, and hardware or software development. Native knees and hip implants were most commonly assessed. The most researched conditions were osteoarthritis, implant loosening or squeaking in vivo and structural damage of implants in vitro. Conclusion: in recent years, AE monitoring showed potential of becoming a useful diagnostic tool for lower limb pathologies. However, further research is needed to refine the existing methods and assess their feasibility in early diagnostics. Significance: The current state of research on AE monitoring for hip and knee joint assessment is described and future research directions are identified.

pass filters are also employed to eliminate unwanted noise, such as sensor and electrical circuit noise, electromagnetic interference, background and technological acoustics noise. Whereas the conventionally used term «acoustic» is applied to sonic waves within the range of human hearing, «acoustic emission» refers to high frequency elastic waves in solids, and the filters' bandwidths can range from several kHz and up to 1 MHz [6]. Following filtering, analog AE signals can then be converted into digital form and be post-processed. These components can be physically incorporated in to a monitoring system in functional blocks (Fig. 1, top) in various order, for example, the main amplifier, filters and analog-to-digital converter can be incorporated in a single data acquisition device (Fig. 1, bottom left), while the pre-amplifier can be embedded in the sensor (Fig.1, bottom right).
A variety of different approaches for signal analysis exist, but the most frequently used method is based on the registration of AE signals that exceed a preset or a floating amplitude threshold. Registered events are referred to as hits, and their characteristics, such as number of hits (counts), duration, rise and fall times, and measured area under the rectified signal envelope (MARSE) are commonly measured (Fig. 2). Specific values of such signal parameters can be an indication of existing defects in the material or formation of microcracks. More advanced methods of AE signal analysis include, for example, wavelet transform, moment tensor and 3D finite element analysis [1]. AE monitoring can be used for the investigation of the behavior of materials and structures under loads, wear and friction, phase transformation, stress corrosion and other material phenomena [1]. The fact that AE monitoring is noninvasive, without harmful side effects associated with radiation exposure, adds to its application in the medical field and particularly in orthopedics. Early works in the application of AE monitoring in orthopedics date back to the midseventies, in topics such as the study of AEs in bones during stretching and re-stressing [7], [8], bone fracture healing in dogs, and osteoporotic bone microstructure and implant behavior [9]. Even though AE monitoring in applied medicine has not been used as extensively as it is in industry [5], a number of studies recently indicated its utility in the assessment of joints due to cartilage and bone deterioration e.g. [10], [11], ligament rupture e.g. [12], [13] and implant loosening e.g. [14]. This renewed interest in medical AE monitoring for non-invasive diagnostics is associated with prevalence of degenerative joint disorders and the increased rate of joint replacement surgeries in recent decades; for instance, it is estimated that the lifetime risk of developing knee osteoarthritis is at 45% in the USA [15], while the risk of undergoing total knee arthroplasty (TKA) or total hip arthroplasty (THA) replacement is at 10% in the UK [16]. For example, as indicated by Kremers et al. [17], only in USA the prevalence of total hip and total knee replacement was 0.83% and 1.52% in 2010, accounting for more than seven million individuals living with artificial hips and knees. Moreover the rate of total joint replacements is likely to continue to increase in the coming decades [17], [18].
While several reviews on the topic in a wider context exist [19]- [21], a well-defined and transparent literature search, data charting and interpretation of findings were not attempted before. Considering the diverse nature of the existing literature, a scoping review was deemed as the most appropriate tool to map the research in this area [22], [23]. The research question posed seeks to determine the nature and extent of the current research in the use of AE monitoring for the condition assessment of human lower limb joints and implants. This scoping review aims to identify knowledge gaps and prospective research directions, and outline the challenges in the application of AE techniques in orthopedics and particularly in joint assessment. Up to date studies were analyzed and synthesized to present advancements in the field and the specifics of the experimental set ups, such as sensor placement, type of monitored activities, and the most indicative AE parameters for specific conditions.

II. METHODS
The scoping review protocol was developed following the PRISMA Extension for Scoping Reviews (PRISMA-ScR) guidelines [24] and was registered in the OSF open registries network [25]. A comprehensive literature search was conducted in April 2020, using Embase, PubMed (incl. MEDLINE), and Web of Science electronic databases, and Google Scholar search engine. This set of databases is considered to provide sufficient recall, averaging at 98.3% for systematic reviews [26]. Database recommendations for systematic reviews should also provide sufficient recall for scoping reviews for journal articles and to ensure extensive coverage of the grey literature, Google Scholar search was additionally extended beyond the recommended 200 records for systematic reviews [15] to 600 in this review. While both PubMed and Embase include all MEDLINE records, differences in article indexing between databases may result in dissimilar outcomes, therefore, both databases were included [27]. Considering the review's aims, no limits on date of publication, source or study type were placed. The search was restricted to studies published in languages spoken by researchers (English and Russian). Title, abstract and keywords search was performed in Embase and Web of Science, and an "all fields" search was used for PubMed and Google Scholar.
Only studies focusing on the condition assessment of hip or knee joints and implants by means of AE monitoring were included, while works investigating material properties, such as damage propagation in bioceramics or bone cement, were excluded. Studies using sensors or filter frequencies below 1 kHz were also excluded, as this frequency range refers to vibroarthrography [28]. Search results included articles with the "acoustic emission" phrase, and at least one of the following search terms: knee, hip, bone, joint, ligament, cartilage, implant or prosthesis. Wildcards were used to account for spelling variations (e.g. prosthe*: prosthetic, prosthesis etc.). Frequently used terms in studies of industrial or non-orthopedic applications were excluded from Google searches ( Table I). The first 600 Google Scholar records were sorted by relevance, and irrelevant papers were excluded based on their titles and short summaries.
All extracted records were imported into Mendeley citation manager to identify duplicates, and were then transferred to Rayyan, an online application for conducting systematic reviews [29]. Title and abstract screening were conducted by a single researcher (LK), and eligible articles were full-text screened. Uncertainties considering study selection and data extraction were resolved by consensus and discussion with the other authors. Finally, backward reference search was performed on the identified reviews on the topic, and all relevant studies were additionally included in this review.
As the scope and nature of the available evidence was not fully known in advance, data extraction tables were created as the records were screened and analyzed. During full text scanning, articles were divided according to their topic into four groups (joint assessment, implant assessment, hardware and software description, and reviews) and data-charting forms were developed to determine which variables to extract for each group. A single reviewer (LK) charted the data, and the findings for each category were summarized.
Articles on joint condition assessment were categorized based on the researched medical condition (e.g. osteoarthritis, age related deterioration, anterior cruciate ligament rapture), physical impact (e.g. mechanical load), and the experimental parameters of each study were logged (e.g. number and type of sensors, placement, frequency range, recorded AE parameters, joint movements, number and characteristics of participants). Similarly, implant assessment studies were documented, along with information on the applied loads, researched conditions (e.g. mechanical failure, squeaking, loosening), and the specimens' description for all in vitro studies. Technical characteristics (e.g. number and types of sensors, frequency range, output signal), applications and validation methods, where applicable, were reported for works on hardware and software in joint or implant assessment. Short descriptions of the included reviews, the identified knowledge gaps and indicated perspectives were reported narratively.

A. Literature search
The search identified 1379 articles (Fig. 3). Duplicate studies were excluded, resulting in 1075 records. Titles and abstracts were reviewed as against the study's inclusion and exclusion criteria, identifying 103 articles at this stage. Eight records were excluded as full texts were not available and one additional record was excluded as the text was written in German. Three conference abstracts, letter and two journal articles [30], [31] were excluded due to limited information being available in respect to the charted parameters of this review. Out of 103 articles, 64 were deemed as relevant during full text screening and were kept for charting (Fig. 3). Two additional newly published papers were identified from Google Scholar alerts and added on a later stage. Reference lists from all the selected review papers were scanned and five additional articles were included, totaling in 71.

B. Joint assessment
Twenty-seven records focused on joint assessments (Supplemental materials: Tables III, IV), of which twenty-six assessed the condition of knees, and one evaluated both hip and knee joints [32]. A number of different medical conditions and their relation to AE signals were investigated (Table II):  the most researched one was osteoarthritis (OA) with thirteen identified records, followed by age-related joint deterioration with six works. The effects of juvenile idiopathic arthritis (JIA), meniscal injury and surgery, and past knee injury or pain joint were also explored on a single publication each. AE signals during anterior cruciate ligament (ACL) rupture under strain were described in [12], [13]. Studies also examined AEs in relation to non-pathological joint conditions such as mechanical loads during movement [33], joint friction [34], the consistency of subject's joint acoustical signals between measurements [34], [35] and stride detection using AE during walking [36]. Only five studies conducted experiments ex vivo (Supplemental materials: Table III). Human cadaveric knee specimens with all soft tissues removed except ACL were used in two studies [12], [13]. Studies also examined AEs in relation to non-pathological joint conditions such as mechanical loads during movement [33], joint friction [34], the consistency of subject's joint acoustical signals between measurements [34], [35] and stride detection using AE during walking [36]. Only five studies conducted experiments ex vivo (Supplemental materials: Table III). Human cadaveric knee specimens with all soft tissues removed except ACL were used in two studies [12], [13]. Meniscal tear, imitation of swelling by injecting saline solution and meniscectomy were performed on human cadaver limbs and their relation to AE during joint movement were investigated in [53]. Samples of hip and knee joint cartilage surfaces were used in [32] to evaluate surface roughness, while AE measurements were performed only in vivo, in participants with varying degree of cartilage deterioration. Femurs condyles pressed against polymeric counterparts simulating the tibial plateau were used in [34] to detect AEs caused by joint friction. Sensors with a frequency range of up to 800 kHz were used in joint condition assessment [12], [13] (Supplemental materials: Table III). In the majority of the included studies, off-the-shelf or custom made piezoelectric (PE) sensors were used to obtain AE signals, however, other types of sensors such as MEMS e.g. [35], electret microphones e.g. [52] and accelerometers e.g. [50] were also utilized (Supplemental materials: Table III). Sensors were usually placed on the areas of bony prominences, such as the patella or the tibia and femur condyles. Regular medical tape was commonly used to fixate sensors on the skin, and a suction cup was used in [49]. Medical ultrasound gels [40], [42]- [45], [49], Vaseline [37] or wax [47] were routinely used to ensure optimal acoustic coupling. The number of sensors varied from one to four per knee (Supplemental materials: Table III). A wide variety of sensors' position on the knee was explored in the reviewed studies (Fig. 4). For the hip joint, sensor positioning was less diverse and was limited in the femoral bone prominences [54], greater trochanter area [55], or in the area from the iliac crest to the upper or mid femur in for studies with several using multiple sensors [56]- [59].
Knee angles were commonly measured along with AE, using electrogoniometers (EG) or inertial measurement units (IMUs), while MRI scans and radiography were deployed as a reference method for assessing joint condition (Supplemental materials: Table III). To generate AE in vivo, different movements were performed by participants; sit-to-stand-to-sit (STSTS) and flexion-extension in sitting position (F/E) were used in the majority of the studies, whereas walking and squats or knee-bends were less common (Supplemental materials: Table III). A leg press machine was also employed in one study [33], and a variety of everyday activities including cycling were used in [34]. Considering the AE parameters of interest, authors focused on AE hits and their characteristics (number, amplitude, energy etc.), and waveform analysis (e.g. frequency distribution, signal patterns) (Supplemental materials: Table III).

C. Implant assessment
Implants and their components (e.g. femoral heads) were assessed in 22 works (Supplemental materials: Table V, VI). The majority of the identified studies investigated hip implants, whilst only three articles [61]- [63] described knee assessments. Squeaking of hip implants was the most frequently addressed issue and was investigated in seven works, both in vitro and in vivo. Other conditions included material fatigue and accumulated damage under loads [64] or compression [65], including microcracks formation e.g. [66], wear [67], loosening or debounding of cement-retained implants [14] (Supplemental materials: Table VI).
Cadaver limbs were used in one study [70] and synthetic specimens (fiber glass or other composite artificial bone with implanted joint prosthesis) were used in 12 studies. For in vitro studies sample size did not exceed twenty specimens, whereas in vivo seven out of nine studies (Supplemental materials: Table 5) had more than twenty participants with a maximum of 98 [55]. The recorded and processed AE parameters were similar to those used in joint assessment, focusing on AE hits parameters, whereas two works looked into specific signal waveforms [55] and spectral characteristics, such as primary frequency, frequency content etc. [58] (Supplemental materials: Table V).

D. Works describing hardware and software
Fifteen works described devices and software for AE monitoring in the context of joint assessment (Supplemental materials: Table VII). Four articles [73]- [76] introduced novel AE sensors with thermosensitive elements to enhance the sensors' sensitivity. An audio-visual environment for multimodal AE analysis of the knee's condition was presented in [77]; the system provided a comparative analysis of two joints by using the animated movements of knees as reconstructed from 3D MRI, synchronized sonified AEs and visualized joint contact areas. A software for AE analysis (3DMem) was presented in [78], designed to investigate the cement's microcrack formations in femoral stems and visualize their location and distribution.
A stationary system for joint acoustic analysis was discussed in [40] and wearable devices for the evaluation of the knee's health were described in [35], [79]- [82]. While all the identified devices and sensors were intended for joint condition assessment, some specific applications were mentioned, such as AE source detection [74], tribological condition evaluation and prediction of femur rupture [81], osteoarthritis evaluation [75], [76], injury monitoring [35], [83], and quantifying rehabilitation stage [79]. Only one work specifically designed a system for hip implant condition assessment [56].
The use of adhesive patches and tapes was the main method of sensor attachment; however, in [35] authors suggested a knee "sleeve" design for future devices, whilst bandages were used in [56]. Works describing devices for knee condition assessment suggested the placement of sensors on the patella area (Supplemental materials: Table V), sensors of the device for hip implant condition assessment were placed from iliac crest to upper femur [56]. Knee angles were frequently measured along with AE by means of electrogoniometers in stationary systems [40] or IMUs in wearable solutions. A temperature and a lower-rate electrical bioimpedance measurement were also included in [82] to provide complex knee assessment, including swelling, activity level, and joint angle.
Authors frequently used off-the-shelf data acquisition hardware, such as the AE PCI-2 board (Physical Acoustic), myRIO (National Instruments) and Biopac modules, whereas custom made hardware were employed in [79], [81]. In stationary systems, the JAAS [40] and BoneDias [81] piezoelectric sensors were employed, whereas the wearable solutions presented in [35], [79]- [81] used microphones. Different types of sensors, such as MEMSs and contact electret microphones, were assessed in [83], [84] in order to determine the most suitable option for wearable applications. Filtered, but otherwise unprocessed acoustic signals were found to be the most frequent output of the developed hardware systems, however, four systems additionally recorded other parameters, such as the number of hits [35], [40], [80], [83].

E. Reviews
Ten review papers were identified. A short narrative review [9] was the earliest work touching the subject of nondestructive techniques for the evaluation of implant performance. A later work [85] provided an extensive overview of the general principles of AE monitoring and its application in the analysis of biomaterials, tissues and tissue/biomaterial interfaces, with a particular focus on detection sensitivity, signal analysis, relation of AE signals to microstructural phenomena and failure mechanisms. Three review papers provided descriptions of different methods of human joint monitoring, including AE analysis: a generic view on joint assessment was presented in [86], while more specific applications, such as monitoring of bone-implant interfaces and implant loosening were discussed in [87] and [88], respectively. An overview of potential biomarkers for knee OA was outlined in [10]. Narrative reviews, fully dedicated to acoustic emission techniques in orthopedics, were presented in [19]- [21], with [21] focusing on current and potential uses of AE monitoring in tribological assessments (i.e. joint wear and friction) and [20] on the evaluation of hip replacement constructs. The latest review [89] in the field of hip implant performance prediction by acoustic techniques covered this relatively narrow topic.

IV. DISCUSSION
The reviewed studies in joint assessment showed positive results (Supplemental materials: Table IV), preliminary confirming the feasibility of using AE monitoring for distinguishing conditions such as OA e.g. [60], age-related deterioration e.g. [11], and trauma [53]. However, only one in vivo study [50] compared joints' AEs before and after an intervention (successful treatment of JIA). The majority of the included studies were pilot or small-scale validation and proof-of-concept works, with 16 out of 24 works including 10 or less participants/specimens per researched group. Given the emerging topic of AE in medicine, authors focused on the description of AE signals to specific conditions, thus indicating the potential use of AE monitoring as a diagnostic tool. However, using AE monitoring for diagnosis prior to the active clinical manifestation of the symptoms has yet to be reported.
Knee joints were more frequently researched than hip joints, most likely due to the higher acoustic wave attenuation from the soft tissues that lay between the hip joint and the sensors. The opposite trend was observed in implant condition assessment; since audible squeaking could be a sign of underlying defects in hip implants, seven studies (Supplemental materials: Table VI) investigated this phenomenon and the mechanisms behind it, in an effort to improve implant design. However, further research into the applications of AE monitoring in knee implant assessment could be expected, considering the higher prevalence of TKAs compared to THAs [17]. Contrary to joint assessment, AE monitoring in implants (Supplemental materials: Table VII), particularly in vitro, seems more promising in distinguishing minor changes in implant structure, such as microdamage [65], microcracks [90] and debounding [63]. Experiments in vitro are easily replicated and they can be adopted for newly developed implant designs to predict their performance in vivo and evaluate possible defects, such as cement microcracks [78], [91].
Regarding the hardware development for joint assessments in vivo, researchers either developed stationary diagnostic systems intended to be used in clinical settings, such as the JAAS [40] and BoneDias [81], or suggested potential solutions for wearable versions capable of remote or long-term monitoring e.g. [35]. For the in vitro assessments, the majority of the studies used commercially available AE equipment originally designed for industrial applications, without indicating the need for the development of specific sensors for human studies.

A. Sensors placement, type and fixation
Sensor placement, particularly in vivo, is one of the key factors in the quality of the signal since appropriate positioning reduces movement artifacts and facilitates signal transmission from the source of the AEs to the skin surface. Sensors should be placed on a boney surface, like the patella or the tibiofermoral condyles, minimizing the acoustic wave's attenuation by soft tissues. In [11], the authors compared four sensor positions on the knee and concluded that sensors on the medial tibia condyle offered minimal muscular and dynamic artifacts for STSTS movements. The same sensor position should also be suitable for a range of movements with minimal knee abduction or rotation.
Sensor fixation is also critical, since tape or straps can generate acoustic emissions when deforming during movements. Yet, none of the reviewed works investigated the potential adverse effects of sensor fixation and how to address them. Contact sensors with coupling gels were most often used, since they facilitate acoustic signal transition from the skin surface to the sensor. At the same time contact sensors are quite susceptible to movement artifacts, and non-contact microphones were suggested as an alternative for wearable solutions. A recent study also concluded that non-contact microphones can be successfully used in either silent or loud background settings with sufficient repeatability [80].

B. Loading and movements
During experiments on implants in vitro or ex vivo, mechanical equipment was used to apply loads or to recreate movements naturally occurring in vivo [14], [56], [57]. While providing excellent repeatability, loading equipment can introduce additional vibrations and AEs into the recorded signal, necessitating the presence of appropriate damping [57]- [59]. Manual manipulation was used as a non-vibrating alternative (e.g. [59], [58]), however, it does not necessarily provide better stability and movement repeatability. Even though a wide range of movements can be performed to trigger AE in vivo, no recommendations were made as to which tasks are optimal for clinical assessments. According to [43], the descending deceleration phase of the stand to sit movement is potentially the most discriminative for quantitative analyses and the monitoring of the knee's ageing and condition; however, only the STSTS task was considered in this study. The STSTS was in general the most commonly recorded activity for joint assessment (11 studies out of 27), yet, such movements exploit the participant's own weight as a load, which may be undesirable in joint condition evaluation since AEs and applied loads are correlated [33]. Therefore, evaluation with AEs can lead to false positive results in obese participants, particularly in the frequently researched pathologies, such as in OA, that are associated with obesity [92]. Alternative tasks with controlled loads and standardized movements, such as leg presses or cycling, should be considered.

C. AE parameters and analysis
It is suggested in [56] that tissue attenuation plays a significant role in AE analyses due to the high and low frequency signals being almost non-distinguishable from each other when recorded on the skin surface. In [58], AE signals were present throughout the whole frequency range of the recordings and reaching up to 50kHz, but signal amplitudes were lower for high frequency signals. Even though tissue attenuation was not specifically investigated in knees, fourteen studies (Supplemental materials: Table III) successfully registered signals with the lower band of the frequency range no higher than 35 kHz. Whereas some studies recorded frequencies up to 500 kHz [47], [60], in the majority of the joint assessment studies, frequencies did not exceed 200kHz. In implant assessments in vitro, the sensors' frequencies were significantly higher and reaching up to 2MHz [69] thanks to the absence of soft tissues and the high coupling quality between sensors and testing surface. Also, high signal sampling rates are also required, which cannot always be implemented. Optimal choice of frequency range was not specifically investigated, however squeaking is commonly recorded and analyzed in a lower bandwidth range, whereas damage and crack formation in a higher one.
The number of hits is a widely reported metric in nondestructive testing, but technical parameters such as the hits' threshold and frequency range are specific to each application. While the threshold's value is a determining factor for AE recordings, it is mostly not reported e.g. [45], [61]; however, the amplification of the AE signal in different hardware can be drastically altered, which in turn results in a variety of thresholds. Full reporting on the system's configuration, including levels of amplification, can be useful in determining the optimal parameters and standard procedures for AE monitoring. Among the other frequently reported metrics, the maximum amplitude distribution, percent occurrence and concentrated distribution of AE hits were considered as optimal biomarkers of OA and other conditions [45], [68]. Frequency distribution analysis (e.g., using Fourier transformation) [46], waveform analysis [54], [61], wavelet transformation [58], [69] or cestrum analysis [31] were also employed for AE joint assessment.
AE monitoring in industrial applications is commonly utilized to detect the location of defects or damage, but due to complicated heterogeneous structure of joints and implants, this prospect is still relatively unexplored in clinical AE monitoring. To date, only one group presented experimental studies [64], [66] and software [78] for microcracks location and visualization in synthetic femurs with cemented prosthesis. In the work by [69], AE bursts were also shown to indicate the location of a possible crack. Difficulties in accurately tracing microcracks arise from the complex composite structure of bone and implants that requires multiple sensors and high computational power. Determining the location of microcracks in vivo can be considered to be extremely difficult due to soft tissue attenuation, signal dispersion and discrepancies in joint structures among subjects.

D. AE monitoring limitations and future research directions
Contrary to industrial applications, wave attenuation, dispersion functions and tribological characteristics are considerably more complex in heterogeneous organic structures such as joints, limiting the application of AE monitoring, particularly in vivo. Development of tissue attenuation models and wave propagation functions for complex structures and soft tissues may aid in addressing these issues in future. Furthermore, considering that AE in non-invasive, in vivo monitoring can only be triggered during movement, the problem of motion artifacts becomes significant. Additional studies are necessary to determine optimal sensor design, placement and fixation methods to resolve this issue. In addition, potential biomarkers and associated AE signal parameters, remain one of the most explored topics in AE clinical monitoring, yet they still need refinement and further clarification in the context of specific joint disorders.
The feasibility of AE monitoring to distinguish between healthy and pathological joints is well established, but early diagnosis or the identification of asymptomatic conditions remains unexplored. Cartilage defects location in vivo based on correlations of AE and contact surfaces change during movement can be considered another promising researched direction that can lead to new insights in OA development and substitute to a degree the expensive methods of diagnostics such as MRI.
Considering the analyses in vitro, the effectiveness of AE in implant's condition monitoring was established by multiple works (Supplemental materials: Table VII); however, the AE technique in determining defect locations in implants is still significantly lags behind successful industrial methods. Apart from complex structure of researched objects, additional difficulties in processing are also emerging from the high sampling rate and computational power which are necessary for a successful analysis.

E. Limitations of the review
The present scoping review has some limitations that should be considered when interpreting the results. As opposed to industrial applications, research on AE emission in medical applications has yet to adopt a standardized terminology, thus relevant studies might not have been covered if authors used terms other than «acoustic emission». Additionally, animal studies, studies on assessment of materials for implants or bones, and closely related methods such as vibroarthrography and resonant frequency monitoring were excluded. In addition, the researchers were unable to obtain full texts of several papers. Another limitation arises from the absence of a quality assessment of the included studies; to achieve a high coverage on the topic, all relevant studies were included. However, recommendations, widely used practices and the obtained results of a wide range of studies were analyzed and reported in this review, making this work useful for a wide variety of researchers in the field of clinical AE monitoring.

V. CONCLUSION
This study presents an overview on the existing research on the use of AE techniques for human lower limb joints and implants' assessment in vitro and in vivo, as well as the current and prospective research directions and knowledge gaps in the field. AE monitoring for implant evaluation was first used more than thirty years ago but modern advances in electronics and increasing prevalence of joint disorders renewed the interest in the technique. While implant assessment remains closely related to material testing in industry, a whole new area of in vivo joint and implant diagnostics emerged, which may lead in the future to the early diagnosis of pathologies or to a wide range of applications in orthopedics.
Future research directions might include further investigations of AE propagation mechanisms in soft and bone tissues, development of mathematical models thereof, development of possible biomarkers for a range of joint conditions, design of sensors specific for in vivo applications, and refinement and standardization of AE monitoring procedures.