Theofanis Ganitidis

Theofanis Ganitidis, Maria Athanasiou, Konstantinos Mitsis et al.

Background: The COVID-19 pandemic has highlighted the need for robust diagnostic tools capable of detecting the disease from diverse and evolving data sources. Machine learning models, especially convolutional neural networks (CNNs), have shown promise. However, the dynamic nature of real-world data can lead to model drift, where performance degrades over time as the underlying data distribution changes. Addressing this challenge is crucial to maintaining accuracy and reliability in diagnostic applications. Objective: This study aims to develop a framework that monitors model drift and employs adaptation mechanisms to mitigate performance fluctuations in COVID-19 detection models trained on dynamic audio data. Methods: Two crowd-sourced COVID-19 audio datasets, COVID-19 Sounds and COSWARA, were used. Each was divided into development and post-development periods. A baseline CNN model was trained and evaluated using cough recordings from the development period. Maximum mean discrepancy (MMD) was used to detect changes in data distributions and model performance between periods. Upon detecting drift, retraining was triggered to update the baseline model. Two adaptation approaches were compared: unsupervised domain adaptation (UDA) and active learning (AL). Results: UDA improved balanced accuracy by up to 22% and 24% for the COVID-19 Sounds and COSWARA datasets, respectively. AL yielded even greater improvements, with increases of up to 30% and 60%, respectively. Conclusions: The proposed framework addresses model drift in COVID-19 detection, enabling continuous adaptation to evolving data. This approach ensures sustained model performance, contributing to robust diagnostic tools for COVID-19 and potentially other infectious diseases.

Daphne Tsolissou, Theofanis Ganitidis, Konstantinos Mitsis et al.

Reliable risk assessment for carotid atheromatous disease remains a major clinical challenge, as it requires integrating diverse clinical and imaging information in a manner that is transparent and interpretable to clinicians. This study investigates the potential of state-of-the-art and recent large vision-language models (LVLMs) for multimodal carotid plaque assessment by integrating ultrasound imaging (USI) with structured clinical, demographic, laboratory, and protein biomarker data. A framework that simulates realistic diagnostic scenarios through interview-style question sequences is proposed, comparing a range of open-source LVLMs, including both general-purpose and medically tuned models. Zero-shot experiments reveal that even if they are very powerful, not all LVLMs can accurately identify imaging modality and anatomy, while all of them perform poorly in accurate risk classification. To address this limitation, LLaVa-NeXT-Vicuna is adapted to the ultrasound domain using low-rank adaptation (LoRA), resulting in substantial improvements in stroke risk stratification. The integration of multimodal tabular data in the form of text further enhances specificity and balanced accuracy, yielding competitive performance compared to prior convolutional neural network (CNN) baselines trained on the same dataset. Our findings highlight both the promise and limitations of LVLMs in ultrasound-based cardiovascular risk prediction, underscoring the importance of multimodal integration, model calibration, and domain adaptation for clinical translation.