A Deep-Learning Algorithm to Predict Short-Term Progression to Geographic Atrophy on Spectral-Domain Optical Coherence Tomography.

Importance: The identification of patients at risk of progressing from intermediate age-related macular degeneration (iAMD) to geographic atrophy (GA) is essential for clinical trials aimed at preventing disease progression. DeepGAze is a fully automated and accurate convolutional neural network-based deep learning algorithm for predicting progression from iAMD to GA within 1 year from spectral-domain optical coherence tomography (SD-OCT) scans. Objective: To develop a deep-learning algorithm based on volumetric SD-OCT scans to predict the progression from iAMD to GA during the year following the scan. Design, Setting, and Participants: This retrospective cohort study included participants with iAMD at baseline and who either progressed or did not progress to GA within the subsequent 13 months. Participants were included from centers in 4 US states. Data set 1 included patients from the Age-Related Eye Disease Study 2 AREDS2 (Ancillary Spectral-Domain Optical Coherence Tomography) A2A study (July 2008 to August 2015). Data sets 2 and 3 included patients with imaging taken in routine clinical care at a tertiary referral center and associated satellites between January 2013 and January 2023. The stored imaging data were retrieved for the purpose of this study from July 1, 2022, to February 1, 2023. Data were analyzed from May 2021 to July 2023. Exposure: A position-aware convolutional neural network with proactive pseudointervention was trained and cross-validated on Bioptigen SD-OCT volumes (data set 1) and validated on 2 external data sets comprising Heidelberg Spectralis SD-OCT scans (data sets 2 and 3). Main Outcomes and Measures: Prediction of progression to GA within 13 months was evaluated with area under the receiver-operator characteristic curves (AUROC) as well as area under the precision-recall curve (AUPRC), sensitivity, specificity, positive predictive value, negative predictive value, and accuracy. Results: The study included a total of 417 patients: 316 in data set 1 (mean [SD] age, 74 [8]; 185 [59%] female), 53 in data set 2, (mean [SD] age, 83 [8]; 32 [60%] female), and 48 in data set 3 (mean [SD] age, 81 [8]; 32 [67%] female). The AUROC for prediction of progression from iAMD to GA within 1 year was 0.94 (95% CI, 0.92-0.95; AUPRC, 0.90 [95% CI, 0.85-0.95]; sensitivity, 0.88 [95% CI, 0.84-0.92]; specificity, 0.90 [95% CI, 0.87-0.92]) for data set 1. The addition of expert-annotated SD-OCT features to the model resulted in no improvement compared to the fully autonomous model (AUROC, 0.95; 95% CI, 0.92-0.95; P = .19). On an independent validation data set (data set 2), the model predicted progression to GA with an AUROC of 0.94 (95% CI, 0.91-0.96; AUPRC, 0.92 [0.89-0.94]; sensitivity, 0.91 [95% CI, 0.74-0.98]; specificity, 0.80 [95% CI, 0.63-0.91]). At a high-specificity operating point, simulated clinical trial recruitment was enriched for patients progressing to GA within 1 year by 8.3- to 20.7-fold (data sets 2 and 3). Conclusions and Relevance: The fully automated, position-aware deep-learning algorithm assessed in this study successfully predicted progression from iAMD to GA over a clinically meaningful time frame. The ability to predict imminent GA progression could facilitate clinical trials aimed at preventing the condition and could guide clinical decision-making regarding screening frequency or treatment initiation.

Deep Learning in Dermatology: A Systematic Review of Current Approaches, Outcomes, and Limitations.

Artificial intelligence (AI) has recently made great advances in image classification and malignancy prediction in the field of dermatology. However, understanding the applicability of AI in clinical dermatology practice remains challenging owing to the variability of models, image data, database characteristics, and variable outcome metrics. This systematic review aims to provide a comprehensive overview of dermatology literature using convolutional neural networks. Furthermore, the review summarizes the current landscape of image datasets, transfer learning approaches, challenges, and limitations within current AI literature and current regulatory pathways for approval of models as clinical decision support tools.

A Feasibility Study on Tribological Origins of Knee Acoustic Emissions.

OBJECTIVE: Considering the knee as a fluid-lubricated system, articulating surfaces undergo different lubrication modes and generate joint acoustic emissions (JAEs). The goal of this study is to compare knee biomechanical signals against synchronously recorded joint sounds and assess the hypothesis that JAEs are attributed to tribological origins. METHODS: JAE, electromyography, ground reaction force signals, and motion capture markers were synchronously recorded from ten healthy subjects while performing two-leg and one-leg squat exercises. The biomechanical signals were processed to calculate a tribological parameter, lubrication coefficient, and JAEs were divided into short windows and processed to extract 64-time-frequency features. The lubrication coefficients and JAE features of two-leg squats were used to label the windows and train a classifier that discriminates the knee lubrication modes only based on JAE features. RESULTS: The classifier was used to predict the label of one-leg squat JAE windows and it achieved a high test-accuracy of 84%. The Pearson correlation coefficient between the estimated friction coefficient and predicted JAE scores was 0.83 ± 0.08. Furthermore, the lubrication coefficient threshold, separating two lubrication modes, decreased by half from two-leg to one-leg squats. This result was consistent with tribological changes in the knee load as it was inversely doubled in one-leg squats. SIGNIFICANCE: This study supports the potential use of JAEs as a quantitative biomarker to extract tribological information. Since arthritis and similar disease impact the roughness of the joint cartilage, the use of JAEs could have broad implications for studying joint frictions and monitoring joint structural changes with wearable devices.

Quantifying Asymmetry Between Medial and Lateral Compartment Knee Loading Forces Using Acoustic Emissions.

OBJECTIVE: Osteoarthritis is the most common type of knee arthritis that can be affected by excessive and compressive loads and can affect one or more compartments of the knee: medial, lateral, and patellofemoral. The medial compartment tends to be the most vulnerable to injuries and research suggests that a better understanding of the medial to lateral load distribution conditions could provide insights to the quantitative usage of knee compartments in activities of daily life. METHODS: Prior to study in an osteoarthritic clinical population which may present with various complicating anatomical and physiological changes, we investigate knee acoustical emissions of able-bodied individuals during a varying width squat exercise which simulates loading asymmetries that would typically be seen in this clinical population. To that end, we present a novel method to quantify the directional bias of asymmetry between the medial and lateral compartment knee joint load in healthy individuals by recording knee acoustical emissions and analyzing them using a deep neural network in a subject independent model. We placed four miniature contact microphones on the medial and lateral sides of the patella on both the left and right leg. We compared the handcrafted audio features with the automated features extracted from the convolutional autoencoder which is an unsupervised model that learns the comprehensive representation of the input to determine whether these automated features can better represent the signal's characteristic in regard to the structural asymmetry of the knee joint. The input to the convolutional auto encoder (CAE) is a time-frequency representation and different types of these images such as spectrogram and scalogram are compared. We alsocompared the multi-sensor fusion approach with the performance of a single sensor to determine the robustness of using multiple sensors. RESULTS: Using a representation learning based approach, we developed a subject independent classification model capable of classifying the asymmetry of the medial and lateral joint load across subjects (accuracy = 83%). CONCLUSION: The result indicates that wavelet coherence which is the time-frequency correlation of two signals using a wavelet transform yields the best accuracy. SIGNIFICANCE: These findings suggest that acoustic signals could potentially quantify the direction of medial to lateral load distribution which would broaden the implications for wearable sensing technology for monitoring cartilage health and factors responsible for cartilage breakdown and assessing appropriate rehabilitation exercises without overloading on one side.

Vibration Stimulation as a Non-Invasive Approach to Monitor the Severity of Meniscus Tears.

Musculoskeletal disorders and injuries are one of the most prevalent medical conditions across age groups. Due to a high load-bearing function, the knee is particularly susceptible to injuries such as meniscus tears. Imaging techniques are commonly used to assess meniscus injuries, though this approach suffers from limitations including high cost, need for skilled personnel, and confinement to laboratory or clinical settings. Vibration-based structural monitoring methods in the form of acoustic emission analysis and vibration stimulation have the potential to address the limits associated with current diagnostic technologies. In this study, an active vibration measurement technique is employed to investigate the presence and severity of meniscus tear in cadaver limbs. In a highly controlled ex vivo experimental design, a series of cadaver knees (n =6) were evaluated under an external vibration, and the frequency response of the joint was analyzed to differentiate the intact and affected samples. Four stages of knee integrity were considered: baseline, sham surgery, meniscus tear, and meniscectomy. Analyzing the frequency response of injured legs showed significant changes compared to the baseline and sham stages at selected frequency bandwidths. Furthermore, a qualitative analytical model of the knee was developed based on the Euler-Bernoulli beam theory representing the meniscus tear as a change in the local stiffness of the system. Similar trends in frequency response modulation were observed in the experimental results and analytical model. These findings serve as a foundation for further development of wearable devices for detection and grading of meniscus tear and for improving our understanding of the physiological effects of injuries on the vibration characteristics of the knee. Such systems can also aid in quantifying rehabilitation progress following reconstructive surgery and / or during physical therapy.

Estimating Knee Joint Load Using Acoustic Emissions During Ambulation.

Quantifying joint load in activities of daily life could lead to improvements in mobility for numerous people; however, current methods for assessing joint load are unsuitable for ubiquitous settings. The aim of this study is to demonstrate that joint acoustic emissions contain information to estimate this internal joint load in a potentially wearable implementation. Eleven healthy, able-bodied individuals performed ambulation tasks under varying speed, incline, and loading conditions while joint acoustic emissions and essential gait measures-electromyography, ground reaction forces, and motion capture trajectories-were collected. The gait measures were synthesized using a neuromuscular model to estimate internal joint contact force which was the target variable for subject-specific machine learning models (XGBoost) trained based on spectral, temporal, cepstral, and amplitude-based features of the joint acoustic emissions. The model using joint acoustic emissions significantly outperformed (p < 0.05) the best estimate without the sounds, the subject-specific average load (MAE = 0.31 ± 0.12 BW), for both seen (MAE = 0.08 ± 0.01 BW) and unseen (MAE = 0.21 ± 0.05 BW) conditions. This demonstrates that joint acoustic emissions contain information that correlates to internal joint contact force and that information is consistent such that unique cases can be estimated.

Fit to Burst: Toward Noninvasive Estimation of Achilles Tendon Load Using Burst Vibrations.

OBJECTIVE: Tendons are essential components of the musculoskeletal system and, as with any mechanical structure, can fail under load. Tendon injuries are common and can be debilitating, and research suggests that a better understanding of their loading conditions could help mitigate injury risk and improve rehabilitation. To that end, we present a novel method of noninvasively assessing parameters related to mechanical load in the Achilles tendon using burst vibrations. METHODS: These vibrations, produced by a small vibration motor on the skin superficial to the tendon, are sensed by a skin-mounted accelerometer, which measures the tendon's response to burst excitation under varying tensile load. In this study, twelve healthy subjects performed a variety of everyday tasks designed to expose the Achilles tendon to a range of loading conditions. To approximate the vibration motor-tendon system and provide an explanation for observed changes in tendon response, a 2-degree-of-freedom mechanical systems model was developed. RESULTS: Reliable, characteristic changes in the burst response profile as a function of Achilles tendon tension were observed during all loading tasks. Using a machine learning-based approach, we developed a regression model capable of accurately estimating net ankle moment-which captures general trends in tendon tension-across a range of walking speeds and across subjects (R2 = 0.85). Simulated results of the mechanical model accurately recreated behaviors observed in vivo. Finally, preliminary, proof-of-concept results from a fully wearable system demonstrated trends similar to those observed in experiments conducted using benchtop equipment. CONCLUSION: These findings suggest that an untethered, unobtrusive system can effectively assess tendon loading during activities of daily life. SIGNIFICANCE: Access to such a system would have broad implications for injury recovery and prevention, athletic training, and the study of human movement.

Robust Longitudinal Ankle Edema Assessment Using Wearable Bioimpedance Spectroscopy.

OBJECTIVE: We present a robust methodology for tracking ankle edema longitudinally based on bioimpedance spectroscopy (BIS). METHODS: We designed a miniaturized BIS measurement system and employed a novel calibration method that enables accurate, high-resolution measurements with substantially lower power consumption than conventional approaches. Using this state-of-the-art wearable BIS measurement system, we developed a differential measurement technique for robust assessment of ankle edema. This technique addresses many of the major challenges in longitudinal BIS-based edema assessment, including day-to-day variability in electrode placement, positional/postural variability, and intersubject variability. RESULTS: We first evaluated the hardware in bench-top testing, and determined the error of the bioimpedance measurements to be 0.4 Ω for the real components and 0.54 Ω for the imaginary components with a resolution of 0.2 Ω. We then validated the hardware and differential measurement technique in: 1) an ex vivo, fresh-frozen, cadaveric limb model, and 2) a cohort of 11 human subjects for proof of concept (eight healthy controls and five subjects with recently acquired acute unilateral ankle injury). CONCLUSION: The hardware design, with novel calibration methodology, and differential measurement technique can potentially enable long-term quantification of ankle edema throughout the course of rehabilitation following acute ankle injuries. SIGNIFICANCE: This could lead to better-informed decision making regarding readiness to return to activities and/or tailoring of rehabilitation activities to an individual's changing needs.

Acoustic Emissions as a Non-invasive Biomarker of the Structural Health of the Knee.

The longitudinal assessment of joint health is a long-standing issue in the management of musculoskeletal injuries. The acoustic emissions (AEs) produced by joint articulation could serve as a biomarker for joint health assessment, but their use has been limited by a lack of mechanistic understanding of their creation. In this paper, we investigate that mechanism using an injury model in human lower-limb cadavers, and relate AEs to joint kinematics. Using our custom joint sound recording system, we recorded the AEs from nine cadaver legs in four stages: at baseline, after a sham surgery, after a meniscus tear, and post-meniscectomy. We compare the resulting AEs using their b-values. We then compare joint anatomy/kinematics to the AEs using the X-ray reconstruction of moving morphology (XROMM) technique. After the meniscus tear the number and amplitude of the AE peaks greatly increased from baseline and sham (b-value = 1.33 ± 0.15; p < 0.05). The XROMM analysis showed a close correlation between the minimal inter-joint distances (0.251 ± 0.082 cm during extension, 0.265 ± .003 during flexion, at 145°) and a large increase in the AEs. This work provides key insight into the nature of joint AEs, and details a novel technique and analysis for recording and interpreting these biosignals.

A Glove-Based Form Factor for Collecting Joint Acoustic Emissions: Design and Validation.

Sounds produced by the articulation of joints have been shown to contain information characteristic of underlying joint health, morphology, and loading. In this work, we explore the use of a novel form factor for non-invasively acquiring acoustic/vibrational signals from the knee joint: an instrumented glove with a fingertip-mounted accelerometer. We validated the glove-based approach by comparing it to conventional mounting techniques (tape and foam microphone pads) in an experimental framework previously shown to reliably alter healthy knee joint sounds (vertical leg press). Measurements from healthy subjects (N = 11) in this proof-of-concept study demonstrated a highly consistent, monotonic, and significant (p < 0.01) increase in low-frequency signal root-mean-squared (RMS) amplitude-a straightforward metric relating to joint grinding loudness-with increasing vertical load across all three techniques. This finding suggests that a glove-based approach is a suitable alternative for collecting joint sounds that eliminates the need for consumables like tape and the interface noise associated with them.

Unobtrusive Heartbeat Detection from Mice Using Sensors Embedded in the Nest.

Unobtrusive monitoring of physio-behavioral variables from animals can minimize variability in preclinical research and thereby maximize the potential for clinical translation. In this paper, we present the design, implementation, and validation of an instrumented nest providing continuous recordings of seismocardiogram (SCG) signals and skin temperature. SCG represents the chest-wall vibrations associated with the heartbeat, and can potentially provide a measure by which individual heartbeats can be detected without the need for electrodes or implantable devices. A non-contact electric field sensor placed in proximity to the animal in the nest was also used to detect respiratory dynamics. The setup was tested with a total of six anesthetized mice. To understand the effects of mouse positioning within the nest on signal quality, the error in heartbeat detection at different positions of the sensor on the body was quantified, with a simultaneously-obtained electrocardiogram (ECG) as the reference standard. At the optimal placement determined with this approach, multiple perturbations were performed such as pinching, changing ambient temperature, and norepinephrine injection to modulate physiology and assess measurement capability. Heartbeat intervals obtained from the ECG and SCG during the perturbations were correlated (R2=0.82) and were in agreement according to Bland-Altman methods (bias: 0.006ms, 95% confidence interval: [-3.79, 3.78]ms) suggesting that SCG can be reliably used for unobtrusive heartbeat detection. Accordingly, the setup can provide a means by which individual heartbeats - and thereby heart rate and heart rate variability indices - can be quantified without the need for any sensors to be attached to the body of the animal.

Quantifying the Effects of Increasing Mechanical Stress on Knee Acoustical Emissions Using Unsupervised Graph Mining.

In this paper, we investigate the effects of increasing mechanical stress on the knee joints by recording knee acoustical emissions and analyze them using an unsupervised graph mining algorithm. We placed miniature contact microphones on four different locations: on the lateral and medial sides of the patella and superficial to the lateral and medial meniscus. We extracted audio features in both time and frequency domains from the acoustical signals and calculated the graph community factor (GCF): an index of heterogeneity (variation) in the sounds due to different loading conditions enforced on the knee. To determine the GCF, a k-nearest neighbor graph was constructed and an Infomap community detection algorithm was used to extract all potential clusters within the graph-the number of detected communities were then quantified with GCF. Measurements from 12 healthy subjects showed that the GCF increased monotonically and significantly with vertical loading forces (mean GCF for no load = 30 and mean GCF for maximum load [body weight] = 39). This suggests that the increased complexity of the emitted sounds is related to the increased forces on the joint. In addition, microphones placed on the medial side of the patella and superficial to the lateral meniscus produced the most variation in the joint sounds. This information can be used to determine the optimal location for the microphones to obtain acoustical emissions with greatest sensitivity to loading. In future work, joint loading quantification based on acoustical emissions and derived GCF can be used for assessing cumulative knee usage and loading during activities, for example for patients rehabilitating knee injuries.

b-Value: A Potential Biomarker for Assessing Knee-Joint Health Using Acoustical Emission Sensing.

This paper explores the novel application of an automated b-value extraction algorithm for the interpretation of sounds produced by the knee joint during movement. Acoustical emissions were recorded from a total of eight subjects with acute knee injuries a first time, within one week of the injury, then a second time, four to six months following corrective surgery and rehabilitation. The data were collected from each subject using miniature electret microphones placed on the medial and lateral side of the patella during knee flexion and extension exercises. From the acoustical signals measured from each subject, we computed the b-value using the modified Gutenberg-Ritcher equation which is widely used in seismology. The b-value increased for each subject's injured knee from immediately following the injury to several months post recovery. (mean b-value: 1.46 ± 0.35 [injured] and 1.92 ± 0.21 [post-surgery and recovery], p < 0.01). In addition, we compared this analysis technique against an unsupervised machine learning algorithm from our previous work and found that the b-value metric can be as effective to differentiate changes in the joint sounds as our prior approach while requiring less computational time and complexity - both of which are preferable for future integration of this technology into a wearable system.

Wearable Vector Electrical Bioimpedance System to Assess Knee Joint Health.

OBJECTIVE: We designed and validated a portable electrical bioimpedance (EBI) system to quantify knee joint health. METHODS: Five separate experiments were performed to demonstrate the: 1) ability of the EBI system to assess knee injury and recovery; 2) interday variability of knee EBI measurements; 3) sensitivity of the system to small changes in interstitial fluid volume; 4) reducing the error of EBI measurements using acceleration signals; and 5) use of the system with dry electrodes integrated to a wearable knee wrap. RESULTS: 1) The absolute difference in resistance ( R) and reactance (X) from the left to the right knee was able to distinguish injured and healthy knees (p < 0.05); the absolute difference in R decreased significantly (p < 0.05) in injured subjects following rehabilitation. 2) The average interday variability (standard deviation) of the absolute difference in knee R was 2.5 Ω and for X was 1.2 Ω. 3) Local heating/cooling resulted in a significant decrease/increase in knee R (p < 0.01). 4) The proposed subject position detection algorithm achieved 97.4% leave-one subject out cross-validated accuracy and 98.2% precision in detecting when the subject is in the correct position to take measurements. 5) Linear regression between the knee R and X measured using the wet electrodes and the designed wearable knee wrap were highly correlated ( R2 = 0.8 and 0.9, respectively). CONCLUSION: This study demonstrates the use of wearable EBI measurements in monitoring knee joint health. SIGNIFICANCE: The proposed wearable system has the potential for assessing knee joint health outside the clinic/lab and help guide rehabilitation.

Quantifying the Consistency of Wearable Knee Acoustical Emission Measurements During Complex Motions.

Knee-joint sounds could potentially be used to noninvasively probe the physical and/or physiological changes in the knee associated with rehabilitation following acute injury. In this paper, a system and methods for investigating the consistency of knee-joint sounds during complex motions in silent and loud background settings are presented. The wearable hardware component of the system consists of a microelectromechanical systems microphone and inertial rate sensors interfaced with a field programmable gate array-based real-time processor to capture knee-joint sound and angle information during three types of motion: flexion-extension (FE), sit-to-stand (SS), and walking (W) tasks. The data were post-processed to extract high-frequency and short-duration joint sounds (clicks) with particular waveform signatures. Such clicks were extracted in the presence of three different sources of interference: background, stepping, and rubbing noise. A histogram-vector Vn(→) was generated from the clicks in a motion-cycle n, where the bin range was 10°. The Euclidean distance between a vector and the arithmetic mean Vav(→) of all vectors in a recording normalized by the Vav(→) is used as a consistency metric dn. Measurements from eight healthy subjects performing FE, SS, and W show that the mean (of mean) consistency metric for all subjects during SS (µ [ µ (dn)] = 0.72 in silent, 0.85 in loud) is smaller compared with the FE (µ [ µ (dn)] = 1.02 in silent, 0.95 in loud) and W ( µ [ µ (dn)] = 0.94 in silent, 0.97 in loud) exercises, thereby implying more consistent click-generation during SS compared with the FE and W. Knee-joint sounds from one subject performing FE during five consecutive work-days (µ [ µ (dn) = 0.72) and five different times of a day (µ [ µ (dn) = 0.73) suggests high consistency of the clicks on different days and throughout a day. This work represents the first time, to the best of our knowledge, that joint sound consistency has been quantified in ambulatory subjects performing every-day activities (e.g., SS, walking). Moreover, it is demonstrated that noise inherent with joint-sound recordings during complex motions in uncontrolled settings does not prevent joint-sound-features from being detected successfully.