Search Results

Medical HAL Lower Limb Type orthotically fits to the lower limbs and trunk; HAL is a gait training device intended to temporarily help improve ambulation upon completion of the HAL gait training intervention. HAL must be used with a Body Weight Support system. HAL is not intended for sports or stair climbing. HAL gait training is intended to be used in conjunction with regular physiotherapy.

The device is intended for individuals with:

• spinal cord iniurv at levels C4 to L5 (ASIA C. ASIA D) and T11 to L5 (ASIA A with Zones of Partial Preservation, ASIA B);

• post stroke paresis

• paraplegia due to progressive neuromuscular diseases (spinal muscular atrophy, spinal and bulbar muscular atrophy, amyotrophic lateral sclerosis, Charcot-Marie-Tooth disease, distal muscular dystrophy, inclusion body myositis, congenital myopathy, muscular dystrophy)

-cerebral palsy and are 12 years or older

-spastic paraplegia caused by either HTLV-1 Associated Myelopathy (HAM) or hereditary spastic paraplegia (HSP)

who exhibit sufficient residual motor and movement-related functions of the hip and knee to trigger and control HAL.

In preparation for HAL gait training, the controller can be used while the exoskeleton is not donned to provide biofeedback training through the visualization of surface electromyography bioelectrical signals recorded.

HAL is intended to be used inside healthcare facilities while under trained medical supervision in accordance with the user assessment and training certification program.

Medical HAL Lower Limb Type is a battery powered bi-lateral ower extremity exoskeleton that provides assistive torque at the knee and hip joints for gait training. HAL is comprised of a controller, a main unit, and sensor shoes in 30 size variations (variation same as predicate: 3 different leg lengths, 2 different leg lengths, 2 different waist widths >> total 24. New size variation: 3 different leg configurations, 1 leg lengths, 2 different waist widths >> total 6) and weighs ~9.5 kg (21 lbs). The main difference between the Model ML05 and ML07 is the leglengths. ML05 has S.M, L, XL sizes, while ML07 has 2S sizes. The device uses legally marketed electrodes (up to 18 electrodes) to record surface electromyography bioelectrical signals that are processed using a propriety signal processing algorithm. The propriety processing algorithm allows the detect surface electromyography bioelectrical signals to control the HAL device in CVC mode and provide visualization of the surface electromyography bioelectrical signals during biofeedback training. The assistive torque can be adjusted using three parameters: sensitivity level, torque tuner, and balance tuner. The device can also provide two additional modes: Cybernic Autonomous Control (CAC) mode and Cybernic Impedance Control (ClC) mode. CAC mode provides assistive torque leg trajectories based on postural cues and sensor shoe measurements. CC mode provides torque to compensate for frictional resistance of the motor based on joint motion. CIC mode does not provide torque assistance for dictating joint trajectories. A trained medical professional (i.e., physical therapist, etc.) can configure, operate, and monitor the device during gait training to make adjustments as needed.

Patients must exhibit sufficient residual motor and movement-related functions of the hip and knee to trigger and control HAL. The patient must be supported by a Body Weight Support (BWS) system before and during device use. The BWS must not be detached from the patient before doffing this device. HAL is not intended to provide sit-stand or stand-sit movements. HAL is capable of gait speeds up to approximately 2 km/hour on level ground. HAL is not intended for sports or stairclimbing.

In preparation to using HAL, the controller can be used while the exoskeleton is not donned to provide biofeedback training through the visualization of surface electromyography bioelectrical signals recorded. HAL is intended to be used in conjunction with regular physiotherapy. HAL is intended to be used inside a medical facility under the supervision of trained medical professionals who have successfully completed the HAL training program.

The provided text, a 510(k) summary for the Medical HAL Lower Limb Type (HAL-ML), describes the device, its intended use, and its equivalence to a predicate device (HAL for Medical Use (Lower Limb Type), K201559). It primarily focuses on regulatory approval and equivalence, particularly regarding the expansion of indications for use to include Cerebral Palsy and Spastic Paraplegia.

While the document references "clinical data to support the safety and efficacy" and "clinical evaluation procedure," it does not provide a detailed breakdown of acceptance criteria or the specific study results proving the device meets those criteria in the format requested. It states that the "nonclinical and clinical tests submitted demonstrate that the device is as safe and as effective, and performs as well as the legally marketed device cleared as K201559." However, it does not offer the granular information needed to fulfill all aspects of your request (e.g., specific performance metrics, sample sizes for test sets, expert qualifications, or MRMC study details).

Therefore, based only on the provided text, I can infer some information regarding the clinical evaluation but cannot fully populate the table or answer all sub-questions as the detailed study design, acceptance criteria with numerical performance data, and other specifics are not disclosed in this regulatory summary.

Here's an attempt to answer your questions based on the available information, with clear indications where the information is not provided in the text:

1. A table of acceptance criteria and the reported device performance

The document does not specify quantitative acceptance criteria or report specific performance metrics for the efficacy of the device in a table format. It broadly states that the clinical evaluation "results are sufficient to support the claims identified in the Indications for Use for this submission" and that the device is "sufficiently safe".

2. Sample sized used for the test set and the data provenance (e.g. country of origin of the data, retrospective or prospective)

• Sample Size for Test Set: Not provided. The document mentions clinical evaluations for five indication groups but does not state the number of subjects in these evaluations.
• Data Provenance: Not provided. The country of origin of the data (e.g., Japan, where the manufacturer is located) and whether the studies were retrospective or prospective are not mentioned.

3. Number of experts used to establish the ground truth for the test set and the qualifications of those experts (e.g. radiologist with 10 years of experience)

This information is not provided in the document. Ground truth for a device like HAL-ML would likely refer to clinical outcomes or functional improvements, which are assessed by medical professionals during the study, rather than "experts establishing ground truth" in the same way it might apply to an imaging AI algorithm. The document mentions "trained medical professionals (i.e., physical therapist, etc.)" configure, operate, and monitor the device, but not their specific role in establishing ground truth for a study.

4. Adjudication method (e.g. 2+1, 3+1, none) for the test set

This information is not provided.

5. If a multi-reader multi-case (MRMC) comparative effectiveness study was done, If so, what was the effect size of how much human readers improve with AI vs without AI assistance

An MRMC study is typically performed for diagnostic imaging devices where human readers interpret medical images. This type of study is not applicable to the Medical HAL Lower Limb Type, which is a gait training device. Therefore, no information on MRMC studies or human reader improvement with AI assistance is present or relevant here.

6. If a standalone (i.e. algorithm only without human-in-the-loop performance) was done

The device is a physical exoskeleton used for gait training, highly dependent on human interaction (patient and trained medical professional). It's not an algorithm-only device. The "propriety processing algorithm" processes sEMG signals to control the device, but its performance is intrinsically tied to the Human-in-the-loop interaction for gait training. Therefore, a "standalone algorithm only" performance study in the typical AI sense is not relevant or described.

7. The type of ground truth used (expert consensus, pathology, outcomes data, etc.)

The document states "The safety and effectiveness of the subject device is demonstrated through the following clinical evaluation procedure for each of the 5 indication groups... The evaluation results are sufficient to support the claims identified in the Indications for Use." This strongly implies that the ground truth would be based on clinical outcomes data related to ambulation improvement, safety, and effectiveness in the specified patient populations. However, the specific metrics or "ground truth" definitions (e.g., specific scores on mobility scales) are not detailed.

8. The sample size for the training set

The document describes clinical evaluation for the safety and effectiveness of the device as a whole. It does not mention a "training set" in the context of an AI/ML model for which a distinct training set would be used. The "propriety processing algorithm" is part of the device's functionality, but the document does not provide details about its development, including specific training set sizes if machine learning were used this way.

9. How the ground truth for the training set was established

As there is no mention of a "training set" in the context of an AI/ML model with externally established ground truth for training purposes, this information is not provided. The algorithm processes sEMG signals to control the device, which is an engineering function, not necessarily a machine learning model that requires a distinct "training set ground truth" in the way a diagnostic AI would.

In summary, the provided 510(k) summary serves as a regulatory document for substantial equivalence, not a detailed scientific publication of clinical trial results. It confirms that clinical evaluations were performed to support the expanded indications but does not provide the granular data, methodology, or specific acceptance criteria and performance statistics that you've requested beyond a general statement of safety and effectiveness.

Ask a specific question about this device

HAL for Medical Use (Lower Limb Type) orthotically fits to the lower limbs and trunk;

HAL is a gait training device intended to temporarily help improve ambulation upon completion of the HAL gait training intervention. HAL must be used with a Body Weight Support system. HAL is not intended for sports or stair climbing. HAL gait training is intended to be used in conjunction with regular physiotherapy.

The device is intended for individuals with:

• spinal cord injury at levels C4 to L5 (ASIA C, ASIA D) and T11 to L5 (ASIA A with Zones of Partial Preservation, ASIA B);

• post stroke paresis

• paraplegia due to progressive neuromuscular diseases (spinal muscular atrophy, spinal and bulbar muscular atrophy, amyotrophic lateral sclerosis, Charcot-Marie-Tooth disease, distal muscular dystrophy, inclusion body myositis, congenital myopathy, muscular dystrophy) who exhibit sufficient residual motor and movement-related functions of the hip and knee to trigger and control HAL

In preparation for HAL gait training, the controller can be used while the exoskeleton is not donned to provide biofeedback training through the visualization of surface electromyography bioelectrical signals recorded.

HAL is intended to be used inside medical facilities while under trained medical supervision in accordance with the user assessment and training certification program

HAL for Medical Use (Lower Limb Type) is a battery powered lower extremity exoskeleton that provides assistive torque at the knee and hip joints for gait training. HAL is comprised of a controller, a main unit, and sensor shoes. The device comes in 8 size variations (4 different leg lengths and 2 different hip widths) for each of the 3 configuration types (doubleleg, right-leg, and left-leg) and weighs ~14 kg (30 lbs). The device uses legally marketed cutaneous electrodes (up to 18 electrodes) to record surface electromyography bioelectrical signals of the hip and knee extensor and flexor muscles when the device is used in Cybernic Voluntary Control (CVC) mode. This mode provides assistive torque at the corresponding ioint (e.g., hip or knee) using surface electromyography bioelectrical signals that are processed using a propriety signal processing algorithm. The propriety processing algorithm allows the device to detect surface electromyography bioelectrical signals to control the HAL device in CVC mode and provide visualization of the surface electromyography bioelectrical signals during biofeedback training. The assistive torque can be adjusted using three parameters: sensitivity level, torque turner, and balance turner. The device can also provide two additional modes: Cybernic Autonomous Control (CAC) mode and Cybernic Impedance Control (CIC) mode. CAC mode provides assistive torque leg trajectories based on postural cues and sensor shoe measurements. CIC mode provides torque to compensate for frictional resistance of the motor based on joint motion. CIC mode does not provide torque assistance for dictating joint trajectories. A trained medical professional (i.e., physician, physical therapist, etc.) can configure, operate, and monitor the device during gait training to make adjustments as needed.

The provided document is a 510(k) Summary for the HAL for Medical Use (Lower Limb Type) device. It describes the device, its intended use, and substantial equivalence to a predicate device (K171909). The document focuses on demonstrating safety and effectiveness, particularly for new patient populations.

It's important to note that this document is an FDA 510(k) summary, which typically presents summarized findings rather than a detailed breakdown of all study methodologies. Therefore, some specific details for each point requested might not be explicitly stated or might require inference from the provided text.

Here's an analysis of the provided information against your requested points:

Acceptance Criteria and Device Performance

The acceptance criteria are not explicitly stated as distinct numerical targets for each performance metric in a single table. Instead, the document demonstrates meeting acceptance criteria through compliance with recognized standards, successful bench testing, and consistent or improved clinical outcomes compared to baseline or control groups across various studies. The "results" sections for non-clinical and clinical data effectively serve as proof of meeting implicit or explicit acceptance criteria related to safety, functionality, and efficacy.

Table of Acceptance Criteria and Reported Device Performance:

Study Details:

This device is a gait training device (exoskeleton), not an AI/imaging device, so many of the requested points related to AI model evaluation, ground truth establishment by experts for image data, MRMC studies, or training/test set sample sizes for an AI algorithm are not directly applicable in the typical sense for this device. The clinical "studies" referred to are more clinical trials or observational studies on human subjects, to demonstrate the effectiveness of the physical device in improving ambulation.

However, I will extract relevant information based on the typical interpretation for evaluating a medical device's performance, applying it to the context of a physical intervention device.

1. A table of acceptance criteria and the reported device performance: refer to the table above.

2. Sample sizes used for the test set and the data provenance:

   • Spinal Cord Injury (SCI) Group (Effectiveness):

     • Sample Sizes: Studies varied from n=8 to n=55. (I-6 studies were assessed for effectiveness). Specific study IDs and their 'n' values:
       • FDA-ID 11 (Aach et al.): n=8
       • FDA-ID 13 (Grasmucke et al.): n=55
       • FDA-ID 17 (Sczesny-Kais): n=11
       • FDA-ID 18 (Jansen et al.): n=21
       • FDA-ID 19 (Jansen et al.): n=8
       • FDA-ID 110 (Puentes et al.): n=12
     • Data Provenance: Not explicitly stated for each study, but the document mentions a "literature search and data held by the manufacturer." The studies are generally chronic SCI patients where spontaneous recovery is not expected, implying these are retrospective analyses of published literature or manufacturer-held data, likely from various international sources (given author names like Aach, Grasmucke, Sczesny-Kais, Jansen, Puentes suggest European/Japanese origins).

   • Stroke Group (Effectiveness):

     • Sample Sizes: Studies varied from n=8 to n=53. (I-14 studies were assessed for effectiveness, categorized by post-stroke stages). Specific study IDs and their 'n' values:
       • I9 (Kawamoto et): n=16 (appears to be a pilot study)
       • I15 (Yoshimoto et): n=18 (for chronic stage)
       • I19 (Tanaka et al.): n=11 (for chronic stage)
       • I20 (Tanaka et al.): n=9 (Chronic, follow-up)
       • I18 (Sczesny-Kais): n=18 (Chronic, crossover RCT)
       • I5 (Nilsson et al.): n=8 (Acute/subacute)
       • I12 (Watanabe et al.): n=22 (Acute/subacute, control group CPT)
       • I14 (Fukuda et al.): n=53 (Acute/subacute)
       • I16 (Tan et al.): n=8 (Acute/subacute)
       • I17 (Puentes et al.): n=11 (Acute/subacute)
       • I11 (Watanabe et al.): n=24 (Acute/subacute, control group CPT)
       • I21 (Yokota et al.): n=37 (Acute stroke rehabilitation)
       • I6 (Yoshikawa et al.): n=16 (End of recovery, comparative study)
       • I13 (Mizukami et al.): n=8 (End of recovery)
     • Data Provenance: Same as SCI group, "literature search and data held by the manufacturer." Given the authors and titles, these are likely retrospective analyses of published literature or manufacturer-held data, likely from various international sources (e.g., Kawamoto, Yoshimoto, Tanaka from Japan; Sczesny-Kais from Europe). One study (I22) and Post-Market Surveillance (PMS) data are specifically mentioned as being from Japan.

   • Progressive Neuromuscular Diseases Group (Effectiveness):

     • Sample Sizes:
       • Literature: 1 case report (I33) with n=3 patients.
       • Clinical Trial: I22, n=24 subjects (investigator-initiated randomized controlled crossover clinical study).
       • Post-Market Surveillance (PMS): n=207 patients (as of November 2019).
     • Data Provenance:
       • Literature: "one published study was assessed."
       • Clinical Trial (I22): Prospective, conducted in Japan, approved by the Ministry of Health, Labour and Welfare of Japan.
       • Post-Market Surveillance: Prospective/Real-World Data, collected over four years in Japan after device approval.

3. Number of experts used to establish the ground truth for the test set and the qualifications of those experts:

   • This question is generally for AI/imaging data. For a physical device, "ground truth" is measured by clinical outcomes (e.g., walking speed, distance).
   • The "ground truth" for the clinical performance (gait function) was established through objective functional tests (e.g., 10MWT - 10-meter walk test, 6MWT - 6-minute walk test, 2MWT - 2-minute walk test), performed without the HAL device. These are standard, quantifiable, and objectively measured clinical endpoints.
   • The "experts" involved would be the trained medical professionals (physicians, physical therapists, etc.) who conducted these assessments as part of the clinical studies. Their specific number or qualifications beyond being "medical professionals" are not detailed in this summary, but it's implied they adhere to clinical trial standards.

4. Adjudication method (e.g. 2+1, 3+1, none) for the test set:

   • This is typically for image interpretation by multiple readers in diagnostic studies. For this device, the "test set" is patient cohorts undergoing a physical intervention, and outcomes are objective measurements.
   • Not Applicable in the sense of radiological adjudication. The outcome measures are performance-based and objectively quantifiable.

5. If a multi reader multi case (MRMC) comparative effectiveness study was done, If so, what was the effect size of how much human readers improve with AI vs without AI assistance:

   • No, an MRMC study was not done. This type of study relates to AI assistance for human readers (e.g., radiologists).
   • The document presents clinical studies comparing HAL intervention with baseline (pre-post) or with conventional physical therapy (control group). These are human subjects assisted by the physical device performing a task (walking), not human readers assisted by AI in interpreting data.
   • Effect Size (where applicable for human patients with device assistance vs. without):
     • SCI (10MWT Speed): Improvements ranged from +0.22 m/s to +0.25 m/s. For time-based measures, improvements were around 35.23s faster (e.g., 70.45s to 35.22s).
     • SCI (6MWT Distance): Improvements ranged from +22.75m to +93.2m.
     • Stroke (10MWT Speed - Chronic): Improvements up to +0.21 m/s.
     • Stroke (10MWT Speed - Acute/Subacute - HAL vs. CPT): The comparative study in the end of recovery stage shows that patients initially treated with HAL reached 61.4 ± 26.6 m/min vs. 50.1 ± 25.0 m/min for CPT, with a significant difference (p<0.05). Another study (I12) showed HAL group 10MWT speed improving by +0.24 m/s (from 0.61 to 0.85 m/s) with statistical significance (p<0.05), while the CPT group's improvement was not significant. The summary states: "significant improvements in the HAL group that were not seen in the control group."
     • Progressive Neuromuscular Diseases (2MWT Distance - Clinical Trial): Treatment effect was -10.066 ± 11.062 for 2MWT in the crossover study (P=0.0369 for the difference, implying HAL improved performance relative to control).
     • Progressive Neuromuscular Diseases (PMS Data): Participants showed about +20% difference from baseline function after 1.5 years despite the progressive nature of their disease.

6. If a standalone (i.e. algorithm only without human-in-the loop performance) was done:

   • Not Applicable in the AI sense. This is a physical device that functions with a human in the loop (the patient wearing it, supervised by medical professionals).
   • The "algorithm" mentioned (propriety signal processing algorithm for sEMG) is part of the device's control system, not a standalone diagnostic AI. Its performance is implicitly validated through the overall device's successful operation in non-clinical tests (e.g., torque output, joint angle, sEMG measurement accuracy) and clinical outcomes.

7. The type of ground truth used (expert concensus, pathology, outcomes data, etc.):

   • For clinical effectiveness: Objective Outcomes Data from standardized functional tests (10MWT speed/time, 6MWT distance, 2MWT distance, TUG test) performed by patients without the device. These are widely accepted and quantifiable measures of ambulation.
   • For clinical safety: Reported Adverse Events (AEs and SAEs) collection (patient reports, clinician observations).
   • For non-clinical performance: Bench test measurements against predefined engineering specifications and standards.

8. The sample size for the training set:

   • This refers to the dataset used to train an AI model. For this physical device, there isn't a "training set" in this sense for a learned AI algorithm that generates the primary output being evaluated.
   • However, if we broadly consider "training" as the development and validation data, that would encompass results from various engineering tests, and potentially earlier developmental clinical work that informed the device design and control algorithms. The document does not provide a specific "training set" sample size for the device's functional logic, as it's not a machine learning model in the typical sense presented for FDA clearance. The control algorithms are described as "proprietary signal processing algorithm".

9. How the ground truth for the training set was established:

   • Not applicable in the context of an AI training set.
   • The device's control logic (e.g., detecting sEMG to trigger movement) is based on fundamental biomechanical principles and signal processing, validated through the non-clinical tests mentioned (e.g., accuracy of sEMG measurement, joint angle measurement). The "ground truth" for calibrating these systems would involve physical measurements, engineering specifications, and physiological data.

Ask a specific question about this device

HAL for Medical Use (Lower Limb Type) orthotically fits to the lower limbs and trunk; the device is intended for individuals with spinal cord injury at levels C4 to L5 (ASIA C, ASIA D) and T11 to L5 (ASIA A with Zones of Partial Preservation, ASIA B), who exhibit sufficient residual motor and movement-related functions of the hip and knee to trigger and control HAL.

HAL is a gait training device intended to temporarily help improve ambulation upon completion of the HAL gait training intervention. HAL must be used with a Body Weight Support system. HAL is not intended for sports or stair climbing. HAL gait training is intended to be used in conjunction with regular physiotherapy.

In preparation for HAL gait training, the controller can be used while the exoskeleton is not donned to provide biofeedback training through the visualization of surface electromyography bioelectrical signals recorded.

HAL is intended to be used inside medical facilities while under trained medical supervision in accordance with the user assessment and training certification program

HAL for Medical Use (Lower Limb Type) is a battery powered bi-lateral lower extremity exoskeleton that provides assistive torque at the knee and hip joints for gait training. HAL is comprised of a controller, a main unit, and sensor shoes. The device comes in 8 size variations (4 different leg lengths and 2 different hip widths) and weighs ~14 kg (30 lbs). The device uses legally marketed cutaneous electrodes (up to 18 electrodes) to record surface electromyography bioelectrical signals of the hip and knee extensor and flexor muscles when the device is used in Cybernic Voluntary Control (CVC) mode. This mode provides assistive torque at the corresponding joint (e.g., hip or knee) using sufface electromyography bioelectrical signals that are processed using a propriety signal processing algorithm. The propriety processing algorithm allows the device to detect surface electromyography bioelectrical signals to control the HAL device in CVC mode and provide visualization of the surface electromyography bioelectrical signals during biofeedback training. The assistive torque can be adjusted using three parameters: sensitivity level. torque turner. and balance turner. The device can also provide two additional modes: Cybernic Autonomous Control (CAC) mode and Cybernic Impedance Control (CIC) mode. CAC mode provides assistive torque leq trajectories based on postural cues and sensor shoe measurements. CIC mode provides torque to compensate for frictional resistance of the motor based on joint motion. CIC mode does not provide torque assistance for dictating joint trajectories. A trained medical professional (i.e., physician, physical therapist, etc.) can configure, operate, and monitor the device during gait training to make adjustments as needed.

Patients must exhibit sufficient residual motor and movement-related functions of the hip and knee to trigger and control HAL. The patient must be supported by a Body Weight Support (BWS) system before donning the device and during device use. The BWS must not be detached from the patient before doffing this device. HAL is not intended to provide sit-stand or stand-sit movements. HAL is capable of gait speeds up to approximately 2 km/hour on level ground. HAL is not intended for sports or stairclimbing.

In preparation to using HAL, the controller can be used while the exoskeleton is not donned to provide biofeedback training through the visualization of surface electromyography bioelectrical signals recorded.

HAL is intended to be used in conjunction with regular physiotherapy. HAL is intended to be used inside a medical facility under the supervision of trained medical professionals who have successfully completed the HAL training program.

Here's a breakdown of the acceptance criteria and study information for the HAL for Medical Use (Lower Limb Type), based on the provided document:

Acceptance Criteria and Device Performance

The document doesn't explicitly state "acceptance criteria" in a separate, quantifiable table for clinical performance endpoints. Instead, it presents the results of two clinical studies and highlights whether the observed improvements are statistically and clinically significant. The key clinical measures used to demonstrate effectiveness are:

Note: The "acceptance criteria" in the table above are inferred from the document's emphasis on demonstrating "statistically significant improvement" and improvement values exceeding "clinically significant" thresholds (MCID) for the 10MWT and 6MWT. For WISCI II, "mean gain of 1.69 levels" with a p-value of <0.001 suggests an accepted improvement. For adverse events, the lack of serious events and resolution of mild events indicates accepted safety.

The document also details numerous non-clinical performance criteria which are met through bench testing to ensure safety and functionality:

• Stopper Strength Test: Conformance maintained after 100 cycles, expected to endure impact.
• Consecutive Landing Test: All 3 samples withstood 3,000,000 cycles without failure, demonstrating sufficient durability.
• Effective Output Test (Torque & Angular Velocity): Output found to meet specifications and be within human tolerance for angular velocity.
• Driving Parts Performance Test: Actual torque output compared to intended torque falls within criteria range.
• Joint angle measurement: Accuracy verified to meet specifications.
• Body trunk absolute angle measurement: Sufficiently detects stable posture for safety and effectiveness.
• Plantar load measurement: Sufficiently detects planting and lifting to determine leg phase for safety and effectiveness.
• Surface Electromyography Bioelectrical signal measurement performance: Accuracy for all measurements (input impedance, common-mode rejection ratio, frequency characteristics) verified to meet specifications.
• Ankle Durability Test: All 3 samples withstood 300,000 impacts (5 years' worth) without failure, demonstrating sufficient durability.

Study Details

2. Sample Size and Data Provenance

DE-01 Clinical Study (Pilot Study):

• Test Set Sample Size: 8 subjects
• Data Provenance: Prospective, conducted at BG University Hospital Bergmannsheil (Germany, inferred from "Bergmannsheil").

DE-02 Clinical Study:

• Test Set Sample Size: 55 subjects
• Data Provenance: Prospective, conducted at BG University Hospital Bergmannsheil (Germany, inferred from "Bergmannsheil").

3. Number of Experts and Qualifications for Ground Truth

The clinical studies involved human assessment of functional outcomes (10MWT, 6MWT, WISCI II). While these are quantitative measures, the execution of the tests and interpretation would be by trained medical professionals. The document states: "The training was supervised by a physiotherapist and a medical doctor." This indicates at least two types of qualified experts were involved in the clinical assessment and potentially in establishing the "ground truth" (i.e., the measured functional scores). Specific years of experience are not provided.

4. Adjudication Method

The document does not explicitly describe an adjudication method for the clinical outcomes. Since the endpoints (10MWT speed, 6MWT distance, WISCI II score) are objective measurements, it is likely that standard clinical protocols were followed for their assessment, which inherently involves some level of consensus or single objective measurement by the supervising medical staff. Formal multi-expert adjudication for "ground truth" as might be seen in image classification is not applicable here.

5. Multi Reader Multi Case (MRMC) Comparative Effectiveness Study

No, a multi-reader multi-case (MRMC) comparative effectiveness study was not explicitly done. The studies were single-arm, uncontrolled interventional studies designed to show improvement within the patient group (before vs. after intervention with HAL) rather than comparing performance against other devices or human readers. The clinical studies focus on the patient's functional improvement without the HAL device after a training intervention, suggesting an "algorithm only without human-in-the-loop performance" in terms of the test subjects' measurement, but the intervention itself is human-in-the-loop (supervised by medical professionals).

6. Standalone (Algorithm Only) Performance

The clinical effectiveness studies (DE-01 and DE-02) do assess a form of standalone performance in the sense that the primary endpoints (10MWT, 6MWT, WISCI II) are measured without wearing the HAL device after the training intervention. This means they are measuring the residual functional improvement in the patient's own ambulation capability after HAL-assisted gait training, rather than the performance of the HAL device itself during use.

The non-clinical bench testing, however, is a form of standalone performance evaluation for the device's mechanical and electrical components and its internal measurement systems (e.g., joint angle sensing, EMG signal measurement accuracy).

7. Type of Ground Truth Used

The ground truth for the clinical studies is functional outcome measures as assessed by standard clinical tests:

• 10 Meter Walk Test (speed)
• 6 Minute Walk Test (distance)
• Walking Index for Spinal Cord Injury II (WISCI II) score

These are objective, quantifiable measures of ambulation capability.

8. Sample Size for the Training Set

The document does not specify a separate "training set" in the context of machine learning model development. The HAL device (Lower Limb Type) is a powered exoskeleton that uses bioelectrical signals and postural cues. While it has a "propriety signal processing algorithm," the document doesn't detail if this algorithm was "trained" on a specific dataset of patients or how large that dataset was. The clinical studies (N=8 and N=55) represent validation/testing of the device's overall effectiveness in improving patient ambulation and its safety.

9. How the Ground Truth for the Training Set was Established

Given that a specific "training set" for an AI/algorithm is not detailed, the ground truth establishment method for such a set isn't provided. The "propriety signal processing algorithm" is mentioned as processing surface electromyography bioelectrical signals to control the device and provide visualization for biofeedback. How this algorithm was developed or optimized (i.e., its "training" process and associated "ground truth") is not disclosed in this regulatory summary.

Ask a specific question about this device