Development of oculomics artificial intelligence for cardiovascular risk factors: A case study in fundus oculomics for HbA1c assessment and clinically relevant considerations for clinicians.

Artificial Intelligence (AI) is transforming healthcare, notably in ophthalmology, where its ability to interpret images and data can significantly enhance disease diagnosis and patient care. Recent developments in oculomics, the integration of ophthalmic features to develop biomarkers for systemic diseases, have demonstrated the potential for providing rapid, non-invasive methods of screening leading to enhance in early detection and improve healthcare quality, particularly in underserved areas. However, the widespread adoption of such AI-based technologies faces challenges primarily related to the trustworthiness of the system. We demonstrate the potential and considerations needed to develop trustworthy AI in oculomics through a pilot study for HbA1c assessment using an AI-based approach. We then discuss various challenges, considerations, and solutions that have been developed for powerful AI technologies in the past in healthcare and subsequently apply these considerations to the oculomics pilot study. Building upon the observations in the study we highlight the challenges and opportunities for advancing trustworthy AI in oculomics. Ultimately, oculomics presents as a powerful and emerging technology in ophthalmology and understanding how to optimize transparency prior to clinical adoption is of utmost importance.

Memory Classifiers for Robust ECG Classification against Physiological Noise.

The development of sophisticated machine learning algorithms has made it possible to detect critical health conditions like cardiac arrhythmia, directly from electrocardiogram (ECG) recordings. Large-scale machine learning models, like deep neural networks, are well known to underperform when subjected to small perturbations which would not pose a challenge to physicians. This is a hurdle that needs to be removed to facilitate wide-scale adoption. We find this to be true even for models trained using data-augmentation schemes.In this paper, we show that using memory classifiers it is possible to attain a boost in robustness using expert-informed features. Memory classifiers combine standard deep neural network training with a domain knowledge-guided similarity metric to boost the robustness of classifiers. We evaluate the performance of the models against naturally occurring physiological perturbations, specifically electrode movement, muscle artifact, and baseline wander noise. Our approach demonstrates improved robustness across all evaluated noises for an average improvement in F1 score of 3.13% compared to models using data augmentation techniques.Clinical relevance- This approach improves the robustness of deep learning methods in safety-critical medical applications.

Electroanatomic Mapping to determine Scar Regions in patients with Atrial Fibrillation.

Left atrial voltage maps are routinely acquired during electroanatomic mapping in patients undergoing catheter ablation for atrial fibrillation (AF). For patients, who have prior catheter ablation when they are in sinus rhythm (SR), the voltage map can be used to identify low voltage areas (LVAs) using a threshold of 0.2 - 0.45 mV. However, such a voltage threshold for maps acquired during AF has not been well established. A prerequisite for defining a voltage threshold is to maximize the topologically matched LVAs between the electroanatomic mapping acquired during AF and SR. This paper demonstrates a new technique to improve the sensitivity and specificity of the matched LVA. This is achieved by computing omni-directional bipolar voltages and applying Gaussian Process Regression based interpolation to derive the AF map. The proposed method is evaluated on a test cohort of 7 male patients, and a total of 46,589 data points were included in analysis. The LVAs in the posterior left atrium and pulmonary vein junction are determined using the standard method and the proposed method. Overall, the proposed method showed patient-specific sensitivity and specificity in matching LVAs of 75.70% and 65.55% for a geometric mean of 70.69%. On average, there was an improvement of 3.00% in the geometric mean, 7.88% improvement in sensitivity, 0.30% improvement in specificity compared to the standard method. The results show that the proposed method is an improvement in matching LVA. This may help develop the voltage threshold to better identify LVA in the left atrium for patients in AF.

In-silico pre-clinical trials for implantable cardioverter defibrillators.

Regulatory authorities require that the safety and efficacy of a new high-risk medical device be proven in a Clinical Trial (CT), in which the effects of the device on a group of patients are compared to the effects of the current standard of care. Phase III trials can run for several years, cost millions of dollars, and expose patients to an unproven device. In this paper, we demonstrate how to use a large group of synthetic patients based on computer modeling to improve the planning of a CT so as to increase the chances of a successful trial for implantable cardioverter defibrillators (ICDs). We developed a computer model of the electrical generation and propagation in the heart. This model was used to generate a large group of heart instances capable of producing episodes of 19 different arrhythmias. We also implemented two arrhythmia detection algorithms from the literature: Rhythm ID from Boston Scientific and PR Logic + Wavelet from Medtronic. Using this setup, we conducted multiple in-silico trials to compare the ability of the two algorithms to appropriately discriminate between potentially fatal Ventricular Tachy-arrhythmias (VT) and nonfatal Supra-Ventricular Tachy-arrhythmias (SVTs). The results of our in-silico trial indicate that Rhythm ID was less able to discriminate between SVT and VT and so may lead to more cases of inappropriate therapy. This corroborates the findings of the Rhythm ID Going Head to Head Trial (RIGHT), a clinical trial that compared the two algorithms in patients. We further demonstrated that the result continues to hold if we vary the distribution of arrhythmias in the synthetic population. We also used the same in-silico cohort to explore the sensitivity of the outcome to different parameter settings of the device algorithms, which is not feasible in a real clinical trial. In-silico trials can provide early insight into the factors which affect the outcome of a CT at a fraction of the cost and duration and without the ethical issues.

Reusable urine glucose sensor based on functionalized graphene oxide conjugated Au electrode with protective layers.

An electrochemical based system with multiple layers coated on a functionalized graphene oxide Au electrode was developed to measure glucose concentration in urine in a more stable way. Two types of gold printed circuit boards were fabricated and graphene oxide was immobilized on their surface by chemical adsorption. Multiple layers, composed of a couple of polymers, were uniformly coated on the surface electrode. This device exhibited higher electrochemical responses against glucose, a greater resistivity in the presence of interferential substances in urine, and durable stabilities for longer periods of time than conventional units. The efficiency in current level according to the order and ratio of solution was evaluated during the immobilization of the layer. The fabricated electrodes were then also evaluated using hyperglycemic clinical samples and compared with the patterns of blood glucose measured with commercially available glucose meters. Our findings show that not only was their pattern similar but this similarity is well correlated.

A wearable healthcare system for cardiac signal monitoring using conductive textile electrodes.

Accurate cardiac signal monitoring feasible for long-term monitoring is important for a practical, cost-effective health monitoring system. In this study, we propose a wearable healthcare system based on conductive fabric-based electrodes allowing monitoring of electrocardiogram (ECG) waveforms and demonstrated the potential for arrhythmia detection using the system. The measurement system uses conductive fabric-based electrodes arranged in a modified bipolar electrode configuration on the chest area of the patient. An adaptive impulse correlation filter (AICF) algorithm and a band pass filter to enable accurate R-peak detection in noisy environments.

Multiple reaction analysis of cancer with different markers using silicon nanowire FET.

In this study, we have used newly developed Silicon nanowire (SiNW) arrays to evaluate their feasibility for the quantification of different markers of interests. We have quantified four different markers of PSA, EGF, IL-6, and VDBP. Each marker showed measurements in the range of 0.184∼17.79 ng/mL (PSA), 10 pg/mL∼10 ng/mL (EGF), 10 pg/mL∼50 ng/mL (IL-6), and 10 pg∼5 ng/mL (VDBP), respectively. For the experiment, we collected 10 different serum samples, 5 prostate cancer patients and 5 breast cancer patients, and measured and compared the resulting signal from the SiNW FET to serum sample from normal patients. As a result, we observed a meaningful pattern of markers associated with each type of cancer. In addition, we have measured the response signal of SiNWs conjugated with Epithelial cell adhesion molecules (EpCAM) markers against tumor cells as they interacted with those markers.

Urine glucose analysis with functionalized graphene oxide as a material for amperometric sensor.

New functionalized graphene oxide (FGO) was systematically coated on the fabricated Au-PCB for the detection of glucose in urine. The electrical response of FGO-Au-PCB exhibited a wide linearity of 1.7∼44.4 mM of glucose levels and a constant variables was less than 3% of the previously performed multiple measurements. The practical application has been demonstrated by measuring the electrical response against glucose in urine samples. In addition, our findings show similar improvement in urine glucose; within each current level, there were significant improvements in urine glucose. Comparison between the urine glucose and blood glucose showed no significant different level from the same subjects.

A data-driven modeling approach to stochastic computation for low-energy biomedical devices.

Low-power devices that can detect clinically relevant correlations in physiologically-complex patient signals can enable systems capable of closed-loop response (e.g., controlled actuation of therapeutic stimulators, continuous recording of disease states, etc.). In ultra-low-power platforms, however, hardware error sources are becoming increasingly limiting. In this paper, we present how data-driven methods, which allow us to accurately model physiological signals, also allow us to effectively model and overcome prominent hardware error sources with nearly no additional overhead. Two applications, EEG-based seizure detection and ECG-based arrhythmia-beat classification, are synthesized to a logic-gate implementation, and two prominent error sources are introduced: (1) SRAM bit-cell errors and (2) logic-gate switching errors ('stuck-at' faults). Using patient data from the CHB-MIT and MIT-BIH databases, performance similar to error-free hardware is achieved even for very high fault rates (up to 0.5 for SRAMs and 7 × 10(-2) for logic) that cause computational bit error rates as high as 50%.

Scalable customization of atrial fibrillation detection in cardiac monitoring devices: increasing detection accuracy through personalized monitoring in large patient populations.

To make it viable for remote monitoring to scale to large patient populations, the accuracy of detectors used to identify patient states of interests must improve. Patient-specific detectors hold the promise of higher accuracy than generic detectors, but the need to train these detectors individually for each patient using expert labeled data limits their scalability. We explore a solution to this challenge in the context of atrial fibrillation (AF) detection. Using patient recordings from the MIT-BIH AF database, we demonstrate the importance of patient specificity and present a scalable method of constructing a personalized detector based on active learning. Using a generic detector having a sensitivity of 76% and a specificity of 57% as its seed, our active learning approach constructs a detector with a sensitivity of 90% and specificity of 85%. This performance approaches that of a patient-specific detector, which has a sensitivity of 94% and specificity of 85%. By selectively choosing examples for training, the active learning approach reduces the amount of expert labeling needed by almost eight fold (compared to the patient-specific detector) while achieving accuracy within 99%.