Multimode fiber (MMF) specklegram sensors have recently drawn significant attention due to the incorporation of machine learning (ML) algorithms in detecting different sensing parameters. Deep learning (DL) techniques provide an efficient way of extracting information from fiber specklegrams, which can be utilized in different sensing applications. Convolutional neural networks (CNN) have proven to be extremely successful in imaging technologies over the last decade. The breakthrough from CNN has instigated new frontiers for applications in different domains. CNNs can automatically learn the variations in MMF specklegrams under different conditions. Besides detecting slight variations in the fiber, CNNs are also insusceptible to environmental noise and fluctuations, thus, making them superior in terms of performance accuracy. They provide a low-cost and simple alternative to extracting information from fiber specklegrams which has piqued the interest of many researchers. In the last few years, there has been a growing number of research articles studying the applicability of DL frameworks in various sensing technologies. In this paper, we present a comprehensive review of such articles that explore the use of ML in different MMF sensing applications, such as bending sensors, endoscopes, tactile or position sensors, and others. The principle of specklegram in MMF, the data generation process, an overview of different DL approaches as well as open challenges and future research directions have also been discussed in this article.

This review presents the current applications of tapered multicore fiber (TMCF)-based optical sensors in single- and multi-parameter sensing. Refractive index (RI)- and temperature-based TMCF sensors are the most widely used candidates. A trench-assisted TMCF sensor demonstrated a sensitivity of 35089.28 nm/RIU within a surrounding RI range of 1.4430–1.4442, the highest reported sensitivity to date among RI sensors. Moreover, TMCF RI-sensor designs for temperature-insensitive operation and in situ temperature compensation have been reported. Additionally, RI-insensitive designs of TMCF temperature sensors were investigated, and a maximum sensitivity of 103.2 pm/oC was obtained. Furthermore, TMCF sensors for measuring strain, curvature, magnetic field, and torsion as well as a localized surface plasmon resonance integrated TMCF biosensor have been reported and discussed. TMCF sensors have shown promise in multi-parameter sensing; however, few reports have been provided in the literature. The potential for multi-parameter sensors is significant due to the higher sensitivity of interference-based sensors to alterations in the effective RI of the optical path. Maximum RI and temperature sensitivities of 5815.50 nm/RIU and 0.1084 nm/oC have been reported for the TMCF configuration as a multi-parameter sensor, respectively. This review serves as a significant resource for researchers who are interested in the area of optical sensing and the design of sensors based on TMCF.

Ischemic stroke, a severe cerebrovascular disease, poses significant health challenges due to its high morbidity and mortality rates. Traditional stroke management relies on direct visual inspection of non contrast MRI (magnetic resonance imaging) or CT (computed tomography) images, which can be affected by equipment artefacts, noise, and the radiologist’s experience. To address these limitations, computer aided diagnostic (CAD) methods, especially deep learning approaches, have gained prominence over the last decade. With their sophisticated data processing abilities, deep learning models are effective for immediate stroke management and prediction. This review explores the impact of deep learning on stroke diagnosis and treatment, highlighting advantages such as reduced treatment duration, improved patient satisfaction, and increased clinician efficiency. It examines five critical areas: timely stroke diagnosis, automated ASPECTS score calculation, large vessel occlusion (LVO) detection, ischemic prognosis, and predicting imaging outcomes. These AI algorithms expedite diagnosis, enhance scoring accuracy, efficiently detect occlusions, and provide accurate prognostic information, aiding disease monitoring and treatment evaluation. Despite advancements, challenges remain, including the need for larger, diverse datasets, improved model interpretability, and extensive external validation. Integrating AI tools into clinical workflows and addressing ethical and regulatory challenges are crucial for broader acceptance and application.