Mitra, Dipankar

Artificial intelligence (AI) is rapidly transforming healthcare through applications in clinical decision support, diagnostic imaging, population health management, and workflow optimization. Despite these advances, real-world deployment continues to expose critical challenges related to safety, bias, transparency, and integration into clinical workflows. Algorithmic bias can exacerbate health disparities, limited explainability may undermine clinician trust, and insufficient validation and post-deployment monitoring can compromise patient safety. Although the World Health Organization (WHO) has established six ethical principles for AI in health, including autonomy, well-being and safety, transparency, accountability, equity, and sustainability, translating these high-level principles into practical and enforceable governance mechanisms remains a persistent challenge. This narrative review synthesizes insights from bioethics, health policy, computer science, and clinical medicine to identify gaps in current AI governance approaches and proposes a lifecycle-aligned governance-by-design framework that operationalizes WHO ethical principles across key stages of the healthcare AI lifecycle, including data collection, model development, validation, deployment, and post-deployment monitoring. The framework integrates concrete governance mechanisms such as consent governance, fairness evaluation, external validation, explainability, clinician oversight, and continuous performance monitoring. Overall, this work advances a practical, lifecycle-integrated approach to AI governance and provides a structured foundation for developing safe, equitable, and trustworthy AI systems in healthcare.

Abstract Pain assessment plays a critical role in clinical decision-making, yet conventional methods rely heavily on subjective self-reports such as the Visual Analog Scale (VAS) or the Numeric Rating Scale (NRS). These approaches fail in scenarios involving infants, sedated patients, unconscious individuals, or those with cognitive or communication impairments. To address this gap, this research proposes PainNet, an AI-driven, real-time facial pain recognition system that integrates deep learning-based object detection with Action Unit (AU)–based pain intensity scoring. The system uses YOLOv8 for rapid facial expression detection and incorporates Py-Feat to extract pain-relevant AUs, enabling computation of the Prkachin and Solomon Pain Intensity (PSPI) score. A balanced dataset comprising 500 images sourced from Delaware, BioVid, and RaFD was curated, annotated, and preprocessed for training. The YOLOv8-based binary pain vs. no-pain classifier achieved a mean Average Precision (mAP) of 0.89 and an F1-score of 0.80 on the test dataset, demonstrating strong performance in diverse environments and facial variations. The design emphasizes computational efficiency by activating AU analysis only when YOLOv8 detects probable pain, making PainNet suitable for continuous monitoring in ICUs, elder-care facilities, and telemedicine systems. This study demonstrates that the fusion of real-time object detection and clinically validated AU scoring can significantly advance automated, objective, and scalable pain assessment.

Background: Interstitial lung disease (ILD) involves numerous chronic pulmonary conditions that damage the lung parenchyma and alveolar interstitium. ILD has overlapping clinical and radiological features with other commonly seen cardiac and respiratory conditions. If not identified and treated in a timely manner, it may lead to irreversible fibrosis and a poor prognosis in the patient. The current diagnostic methods are either invasive or reliant on imaging or specialist interpretation, which can lead to diagnostic delay, increased radiation exposure, and healthcare costs. Lung crackles, often under-recognized as a non-specific feature of ILD, may serve as an important diagnostic clue in identifying not only the early stages of ILD but also its subtypes. This review explores the potential of analyzing the lung sounds in ILD through AI-based auscultation. Objective: To provide a comprehensive analysis of the pathophysiological stages of lung injury in ILD, the specific acoustic features, and the location associated with each ILD subtype and to evaluate the current state-of-the-art non-AI and AI methodologies that are used to diagnose ILD. This review aims to analyze the limitations associated with the current modalities and to envision AI-integrated auscultation as a powerful, cost-effective, non-invasive, radiation-free screening tool for early detection of ILD and its subtypes. Content Overview: The review begins with a detailed analysis of the lung sound pathophysiology, exploring the two-stage mechanism of alveolar epithelial injury and fibrosis formation. Existing hypotheses explaining the mechanism behind crackle production and the role of structural anatomy and surface tension in the generation of pathological lung sounds are examined. A tabulated summary of common ILD subtypes is provided, including their inciting events, pathogenesis, anatomical auscultation locations, and prognostic implications. Current diagnostic modalities for ILD, both non-AI and AI-based, are summarized along with their limitations, emphasizing the need for improved diagnostic tools. Discussion: Existing studies suggest that AI-based auscultation can match or exceed the current modalities in its sensitivity and specificity for detecting ILD-related crackles. Clinicians can identify the specific sound pattern and then correlate it with the ILD subtype and understand the prognosis in real time, thereby providing timely intervention to the patient. Additionally, AI-based auscultation can be used in resource-limited settings and can potentially reduce dependence on pulmonology expertise and radiation-based imaging for monitoring the condition. Conclusions: This literature review highlights the clinical potential of AI-based auscultation for early and accurate diagnoses of ILD. Understanding the associated pathological sounds, biomarkers, and genetic mutations linked to different subtypes opens avenues for future development of non-invasive diagnostic panels for ILD in clinical practice.

Systemic sclerosis-associated pulmonary arterial hypertension (SSc-PAH) is a life-threatening vascular complication of SSc, marked by high morbidity and mortality. Early diagnosis remains a major challenge due to nonspecific symptoms and the limitations of conventional tools such as echocardiography (ECHO), pulmonary function tests (PFTs), and serum biomarkers. This review evaluates the emerging role of artificial intelligence (AI), particularly machine learning (ML) and deep learning (DL), in improving the diagnostic landscape of SSc-PAH. A comprehensive literature search was conducted across PubMed, Scopus, IEEE Xplore, Embase and Google Scholar to identify studies involving AI applications in SSc, pulmonary arterial hypertension (PAH), and their intersection. Evidence indicates that AI models can assist interpretation across modalities, including heart sounds, ECGs, chest X-rays (CXRs), ECHOs, CT pulmonary angiography (CTPA), and omics-based biomarkers. While several models show encouraging diagnostic performance, their accuracy varies by dataset and modality, and most require external validation against right heart catheterization (RHC)-confirmed cohorts. Integrating multimodal data through AI frameworks may enhance early recognition and individualized risk stratification; however, these tools remain exploratory. Future work should emphasize harmonized hemodynamic definitions, transparent validation protocols, and SSc-specific datasets to ensure clinical applicability and reproducibility.

Abstract Medical devices that use electromagnetic fields for diagnostic and therapeutic purposes are rapidly evolving. Understanding the interactions between electromagnetic signals and biological tissue is fundamental to developing safe, novel diagnostic and therapy treatments. This study characterizes the dielectric properties of porcine hearts from 10 male and 10 female animals. Dielectric properties are measured using an open-ended coaxial probe dielectric spectroscopy method over a frequency range of 0.5 to 20 GHz. Scatter parameters (S11) collected from various anatomical cardiac locations (LA, RA, LV, RV, SA, IAS, IVS) are converted into frequency-dependent complex permittivity. The calculated dielectric properties vary regionally within the heart, indicating heterogeneity in tissue properties observed at microwave frequencies.

The human voice is an important medium of communication and expression of feelings or thoughts. Disruption in the regulatory systems of the human voice can be analyzed and used as a diagnostic tool, labeling voice as a potential “biomarker”. Conversational artificial intelligence is at the core of voice-powered technologies, enabling intelligent interactions between machines. Due to its richness and availability, voice can be leveraged for predictive analytics and enhanced healthcare insights. Utilizing this idea, we reviewed artificial intelligence (AI) models that have executed vocal analysis and their outcomes. Recordings undergo extraction of useful vocal features to be analyzed by neural networks and machine learning models. Studies reveal machine learning models to be superior to spectral analysis in dynamically combining the huge amount of data of vocal features. Clinical applications of a vocal biomarker exist in neurological diseases such as Parkinson’s, Alzheimer’s, psychological disorders, DM, CHF, CAD, aspiration, GERD, and pulmonary diseases, including COVID-19. The primary ethical challenge when incorporating voice as a diagnostic tool is that of privacy and security. To eliminate this, encryption methods exist to convert patient-identifiable vocal data into a more secure, private nature. Advancements in AI have expanded the capabilities and future potential of voice as a digital health solution.