
Diatoms are unicellular photosynthetic microalgae characterised by a silica cell wall, the frustule, whose hierarchically nanostructured architecture has motivated their cultivation for biotechnological applications. Frustule morphology depends on the physiological state of the culture during growth, so reproducible production requires continuous monitoring of cell density and pigment composition. Current inline monitoring approaches employ LED sensors originally tuned to chlorophyte pigments, without accounting for the distinctive pigment composition of diatoms – fucoxanthin, chlorophyll c. A further unaddressed limitation of LED-photodiode systems is temperature-induced signal drift, which can obscure the small optical density (OD) changes associated with early-stage culture growth. Here we present a flow-through three-channel LED photocolorimeter with channels at 460, 525, and 625 nm – covering the principal diatom pigment-absorption regions – combined with a biofouling-resistant PTFE cuvette, an OPT101P monolithic detector, and an active temperature-compensation model. Compensation reduced long-term OD variability by 8–57% across the three channels (57% at 460 nm), and photometric linearity was verified against a bench-spectrophotometer reference using haemoglobin solutions. Initial validation on Asterionella formosa suspensions gave a linear OD-concentration response on all channels (R2 > 0.993), with a limit of detection of 337 cells/mL at 460 nm. The instrument is ready for closed-loop bioreactor monitoring of diatom cultures, and the compensation approach is transferable to other LED-photodiode systems limited by thermal drift.
Effective diabetes management relies heavily on non-invasive glucose monitoring coupled with autonomous insulin delivery. This study presents a novel, intelligent closed-loop artificial pancreas integrating non-invasive optical sensing with precise microfluidic drug administration. Photoplethysmography (PPG) waveforms acquired across a cohort of N = 120 subjects were used to predict blood glucose levels via an optimized Extreme Gradient Boosting (XGBoost) model, achieving high predictive accuracy (R2 = 0.89, MAE = 5.4 mg/dL). The predicted glucose concentrations feed into a Mamdani Fuzzy Logic Controller (FLC), which dynamically calculates the required insulin dosage. Physical drug delivery is executed by a microfabricated valveless piezoelectric polydimethylsiloxane (PDMS) micropump. Multi-physics Fluid-Structure Interaction (FSI) simulations performed in COMSOL Multiphysics were validated through physical benchtop testing. Under a simulated 75 g carbohydrate meal disturbance, the closed-loop system successfully restored euglycemia within 85 min without triggering hypoglycemic events. Furthermore, while simulations established a baseline operating voltage of 170 V (0.10 mL/min), the high elasticity of the PDMS membrane enabled the hardware prototype to achieve a stable flow rate of 0.08 mL/min at a safe 30 V driving limit. These findings confirm that integrating optical sensing, machine learning prediction, and compliant microfluidic actuation offers a highly viable, non-invasive pathway toward safe and responsive automated diabetes management.
Accurate liver segmentation in computed tomography imaging is a critical component of diagnostic workflows, surgical planning, and therapy monitoring. Transformer architectures have recently reshaped medical image analysis by providing global contextual modeling capabilities that exceed those of traditional convolutional networks. This review synthesizes Transformer-based liver segmentation models identified through a PRISMA-guided literature search. The included models are systematically categorized according to three principal axes: their architectural integration strategy, the attention mechanism scope, and their inherent transformer dimensionality. Performance comparisons across several benchmarks, LiTS, 3DIRCADb, Synapse, and SLIVER07, demonstrate that contemporary approaches regularly achieve Dice similarity coefficients above 97%. In addition, we analyze training practices, computational complexity, and design trade-offs that influence model performance and clinical applicability. The review concludes by highlighting persistent challenges and discussing promising research avenues toward efficient, generalizable, and clinically deployable Transformer-based liver segmentation systems.
This review (Part III) synthesizes current research on the in vivo toxicity of multimodal upconversion nanoparticles (UCNPs) for biomedical applications, including imaging, photodynamic therapy (PDT), and combination therapies. The aim is to assess the safety of UCNPs in complex formulations combining PDT with chemotherapy, photothermal therapy (PTT), chemodynamic therapy (CDT), immunotherapy, radiotherapy, or cryotherapy. Analysis of numerous in vivo studies in mouse models demonstrates that multimodal UCNP platforms exhibit low systemic toxicity, with no significant damage to major organs (heart, liver, spleen, lungs, kidneys) at therapeutic doses. For mice, safe doses are 10 mg/kg for short-term exposure (12 h) and 50 mg/kg for long-term exposure (120 days). Surface coatings (polyethylene glycol (PEG), silica, cell membranes) and targeting ligands enhance biocompatibility. Multimodal UCNP platforms demonstrate enhanced antitumor efficacy (tumor growth inhibition up to 97.2%, complete eradication in some cases) without increasing toxicity. These findings support the continued development of UCNP-based theranostic agents for precision oncology.
This is the second section of review synthesizes current research on the in vivo toxicity of upconversion nanoparticles (UCNPs) for biomedical applications, including photoinduced drug delivery, photodynamic therapy (PDT), and multimodal therapies. The aim is to assess the safety of UCNPs under near-infrared (NIR) irradiation and in combination with photosensitizers, chemotherapeutics, and other therapeutic modalities. Analysis of numerous in vivo studies in mouse and rat models demonstrates that UCNPs exhibit low dark toxicity and targeted phototoxicity, with no significant damage to major organs (heart, liver, spleen, lungs, kidneys) at therapeutic doses. For mice, safe doses are 10 mg/kg for short-term exposure (48 h and for long-term exposure 40 days). Surface coatings (PEG, silica, MVEMA, cell membranes) enhance biocompatibility and reduce off-target effects. Multimodal UCNP platforms (PDT combined with chemotherapy, photothermal therapy, chemodynamic therapy, immunotherapy, or radiotherapy) demonstrate effective tumor suppression without substantial systemic toxicity. These findings support the continued development of UCNPs as safe and effective theranostic agents for precision oncology.