6 April 2022 in IEEE Access
Diabetic Retinopathy (DR) - a complication developed as a result of heightened blood glucose levels- is deemed as one of the most sight-threatening diseases. Unfortunately, DR screening is manually acquired by an ophthalmologist, a process that can be considered erroneous and timeconsuming. Accordingly, in recent years, automated DR diagnostics have become a focus of research due to the tremendous increase in diabetic patients. Moreover, the recent accomplishments demonstrated by Convolutional Neural Networks (CNN) settle them as state-of-the-art for DR stage identification. This paper proposes a new automatic deep-learning-based approach for severity detection by utilizing a single Color Fundus photograph (CFP). The proposed technique employs DenseNet169’s encoder to construct a visual embedding. Furthermore, Convolutional Block Attention Module (CBAM) is introduced on top of the encoder to reinforce its discriminative power. The model is trained using cross-entropy loss on the Kaggle Asia Pacific Tele-Ophthalmology Society’s (APTOS) dataset. On the binary classification task, we accomplished (97% accuracy - 97% sensitivity - 98.3% specificity - 0.9455, Quadratic Weighted Kappa score (QWK)) compared to the state-of-the-art. Moreover, Our network showed high competency (82% accuracy - 0.888 (QWK)) for severity grading. The major contribution of the proposed framework is that it efficiently grades the severity level of diabetic retinopathy, while reducing the time and space complexity required, which demonstrates it as a promising candidate for autonomous diagnosis.
4 January 2023 in IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control
Tissue Harmonic Imaging (THI) is an invaluable tool in clinical ultrasound owing to its enhanced contrast resolution and reduced reverberation clutter in comparison to fundamental mode imaging. However, harmonic content separation based on high pass filtering suffers from potential contrast degradation or lower axial resolution due to spectral leakage. Whereas nonlinear multi-pulse harmonic imaging schemes, such as amplitude modulation and pulse inversion, suffer from a reduced framerate and comparatively higher motion artifacts due to the necessity of at least two pulse echo acquisitions. To address this problem, we propose a deep-learning-based single-shot harmonic imaging technique capable of generating comparable image quality to pulse amplitude modulation methods, yet at a higher framerate and with fewer motion artifacts. Specifically, an asymmetric convolutional encoder-decoder structure is designed to estimate the combination of the echoes resulting from the half-amplitude transmissions using the echo produced from the full amplitude transmission as input. The echoes were acquired with the checkerboard amplitude modulation technique for training. The model was evaluated across various targets and samples to illustrate generalizability as well as the possibility and impact of transfer learning. Furthermore, for possible interpretability of the network, we investigate if the latent space of the encoder holds information on the nonlinearity parameter of the medium. We demonstrate the ability of the proposed approach to generate harmonic images with a single firing that are comparable to those from a multi-pulse acquisition.
30 March 2026 in European Heart Journal - Digital Health
Aims: Monitoring pulmonary congestion in chronic heart failure (HF) reduces decompensation and hospitalization, but conventional methods such as weight and symptom tracking are often unreliable. As fluid accumulation affects the lungs and vocal tract, subtle voice alterations may serve as a non-invasive signal for early detection of worsening HF. Methods and results: The Voice Analysis for Monitoring Patients with HF trial (VAMP-HF, NCT06566911) prospectively enrolled 104 patients hospitalized with acute decompensated HF (ADHF) across two academic centres in the USA and Germany. Daily voice recordings were collected from admission to discharge, with breathing features extracted from speech and acoustic features from sustained vowels. A machine-learning model was trained to classify recordings as admission-phase vs. discharge-phase using leave-one-patient-out. Patients with clinical deterioration, insufficient audio quality, or short length of stay were excluded. Seventy-nine patients were included in the final dataset. The model classified admission and discharge with an F1-score of 0.83 (95% CI: 0.77–0.90; AUC = 0.90). In patients with higher audio volume (N = 54), performance reached 0.89 (95% CI: 0.82–0.94; AUC = 0.91). When applied to intermediate hospitalization days, model-predicted scores showed progressive increases from admission towards discharge. Performance remained robust irrespective of significant weight loss during hospitalization. Conclusion: In this pilot study, structured voice and breathing analysis discriminated hospitalization phase from admission through discharge in patients with ADHF. This non-invasive approach captured progressive changes during the hospital course and warrants further investigation with concurrent objective congestion markers to establish physiological specificity.
18 June 2026 in European Journal of Heart Failure
1 January 2026 in IEEE Transactions on Ultrasonics
Nonlinear imaging of contrast agents has emerged as a prominent technique in medical ultrasound (US) applications, owing to its enhanced contrast sensitivity and minimized tissue signal interference in comparison to fundamental imaging. This advancement has facilitated the widespread adoption of contrast-enhanced US (CEUS) imaging in various fields, including visualizing vascular architecture within scanned organs. Nevertheless, challenges persist in extracting the nonlinear signal of the contrast agents from the backscattered echo, with conventional methods facing spectral leakage in case of linear filtering, or M-fold reduction in frame rate and increased motion artifacts in case of M-pulse techniques. To address these challenges, we propose a signal-based deep learning framework for single-shot nonlinear contrast imaging (NLCI) in CEUS applications. Here, the network is trained to predict the subsequent pulse echoes required for different multipulse harmonic imaging techniques from a single pulse-echo acquisition. The proposed method yields comparable quality and contrast metrics to conventional M-pulse techniques, including pulse inversion (PI), amplitude modulation (AM), and checkerboard apertures harmonic imaging, across phantoms and in vivo data. This is achieved while attaining an improved framerate by a factor of $M$ , reduced susceptibility to motion artifacts, and lower noise levels.