Open Access

Supervised learning requires a lot of annotated data, which makes the annotation process time-consuming and expensive. Active Learning (AL) offers a promising solution by reducing the number of labeled data needed while maintaining model performance. This work focuses on the application of supervised learning and AL for (named) entity recognition, which is a subdiscipline of Natural Language Processing (NLP). Despite the potential of AL in this area, there is still a limited understanding of the performance of different approaches. We address this gap by conducting a comparative performance analysis with diverse, carefully selected corpora and AL strategies. Thereby, we establish a standardized evaluation setting to ensure reproducibility and consistency across experiments. With our analysis, we discover scenarios where AL provides performance improvements and others where its benefits are limited. In particular, we find that strategies including historical information from the learn ing process and maximizing entity information yield the most significant improvements. Our findings can guide researchers and practitioners in optimizing their annotation efforts.

Industrial digestates from short-fibre residues, generated in paper recycling mills, are driving interest in resource recovery. This study aims to explore their potential for water recovery. Understanding particle dynamics aids in optimizing dewatering for digestate management. The particle size distribution in this study revealed significant fractions: <0.63 μm (6–20%), 0.63–20 μm (38–52%), and >20 μm (11–16%). Pre-treatment with Na4P2O7 and H2O2 enhances settling and lowers total dissolved solids (TDSs) but results in variation of size distribution. Additionally, this study investigates further water reuse in paper mills, focusing on the quality of ultrafiltration (UF) permeate obtained from the digestate of short fibres. UF permeate analysis reveals deviations from freshwater standards in paper mills. Despite effective TS removal, UF permeate falls short of paper mill water standards due to high TDSs, electrical conductivity, and nutrient concentrations, necessitating further downstream treatment with nanofiltration or reverse osmosis. A substantial reduction of permeate flux from 31 to 5 L/(m2·h) over the time indicated fouling and inefficient membrane wash. The silt density index of the UF membrane at 30 min registered 2.1, suggesting potential fouling. Further investigations on optimizing UF operations to enhance permeate flux and exploring alternative UF membranes are required.

Magnetic nanoparticles (MNP) are widely investigated for biomedical applications in diagnostics (e.g. imaging), therapeutics (e.g. hyperthermia) and general biosensing. For all these applications, the MNPs’ unique magnetic relaxation mechanism in an alternating magnetic field (AFM) is stimulated to induce desired effects. Whereas magnetic fluid hyperthermia (MFH) and magnetic particle imaging (MPI) are the most prominent examples for biomedical application, we investigate the relatively new biosensing application of frequency mixing magnetic detection (FMMD) from a fundamental perspective. Generally, we ask how specific MNP parameters (core size, magnetic anisotropy) influence the signal, specifically we predict the most effective MNP core size for signal generation. In FMMD, simultaneously two AFM are applied: a low-frequency magnetic driving field, driving MNP close to saturation, and a high-frequency excitation field that probes MNP susceptibility: . Resulting from the nonlinear magnetization of the MNP, harmonics of both individual incident frequencies as well as intermodulation products of these frequencies are generated. In this work, we present numerical Monte-Carlo(MC)-based simulations of the MNP relaxation process, solving the Landau-Lifshitz-Gilbert (LLG) equation to predict FMMD signals: As Figure 1 shows for the first four intermodulation signals , with , we can clearly see that larger core sizes generally increase the signal intensity. Same trend is predicted by a simple Langevin-function based thermal equilibrium model. Both predictions include a lognormal size distribution. The effect of core size distribution presumably dominates the effect of magnetic anisotropy. The findings are supported by comparison with experimental data and help to identify which MNP are best suited for magnetic biosensing applications using FMMD.

Hyperthermia with the use of magnetic nanoparticles (MNP) is a challenging but most promising approach for cancer therapy. After being magnetically trapped at the tumor site, MNP are heated in alternating magnetic fields (AMF) to approx. 43 °C, which causes tumor cell apoptosis. For an effective and controllable hyperthermia application, two parameters are most important: the amount of internalized MNP in tumor cells and their heating characteristics in AMF. In this study, we evaluated if a sufficient temperature could be achieved by cell internalized MNP heated up in AMF and if cell death could be induced in this way. The heating of pancreatic tumor cell lines MiaPaCa-2 and BxPC-3 loaded with different amounts of selfsynthesized magnetoliposomes nanoparticles (MLs) was measured with a custom-built setup. The MLs consisted of a fluorescent bi-layer of phospholipids and multiple magnetite (Fe3O4) cores with a diameter of (10.0 ± 0.5) nm each. The hydrodynamic diameter of the MLs was (90 ± 5) nm. Cell loading was performed by incubation of tumor cells for up to 24 h at 37 °C in a DMEM cell medium with MLs, which had an iron concentration of 150 μg/mL. Transmission electron microscopy and fluorescence microscopy were used to depict the uptake of MLs into the tumor cells (see Figure). The internalized iron-content per cell was determined with a magnetic particle spectrometer (MPS). After application of AMF for approx. 30 min, cell viability was assessed by clonogenic assay. The cellular uptake of MLs was time-dependent, cell line-specific and saturated: For both MiaPaCa-2 and BxPC-3 cell lines, the MLs cell internalization followed an exponential growth function which saturated after about 24 h cell incubation time at an iron load of (110 ± 6) pg/cell and (30 ± 2) pg/cell, respectively. The time constants of the exponential growth were (7.2 ± 1.4) h and (4.0 ± 0.6) h, respectively. In AMF, cells with the saturated MLs loading reached temperatures of approx. 44 °C and 43.5 °C, which caused the cell survival fraction to drop to approx. one third compared to untreated tumor cells for both MiaPaCa-2 and BxPC-3 cell lines. These results demonstrate the feasibility of hyperthermia in pancreatic cancer treatment by confirming cell death of pancreatic tumor cells at temperatures of approx. 43 °C. Further investigations are planned, which aim for the optimization of MNP dosage in targeting experiments as well as the assessment of incubation times and AMF parameters needed for a successful hyperthermic therapy.

The heat generated by magnetic nanoparticles (MNP) forms the basis of magnetic fluid hyperthermia (MFH) tumor therapy and arises from MNP magnetic moments relaxing in an alternating magnetic field. In physiological environments MNP strongly interact with cells, binding to their membranes as well as internalizing inside lysosomes, which alters the MNP magnetic relaxation. In the present study, we investigate the heating behavior of MNP in-vitro for different binding states and compare it to the heating of trapped MNP in dedicated hydrogels of different mesh size, mimicking different immobilization states. We used iron-oxide MNP (mean core size 10 nm) with a biocompatible phospholipid coating, referred to as magnetoliposomes (ML), for in-vitro studies, and with citric acid coating (CA-MNP) for studies in hydrogels. All samples were subjected to an AMF (40 kA/m, 270 kHz) for 30 min, and from the recorded time-temperature curve, the specific loss power (SLP) value was calculated. In-vitro experiments were performed with L929 cells, which were incubated for 24 h with 225 μg(Fe)/mL ML dispersed in RPMI cell medium. The results of the SLP values were analyzed regarding the internalized ML amount with respect to ML residuals in RMPI medium and compared to fully immobilized ML after freeze-drying (FD): The SLP value of 10.1 % intracellular ML decreased by 20 %, the SLP value of 100 % intracellular ML decreased by 60 % and that of FD-ML decreased by 70 % (Fig 1a). The influence of gradual immobilization of MNP on the heating was investigated by mixing CA-MNP in low-melting agarose and polyacrylamide hydrogels. In agarose and polyacrylamide gels the mean mesh size can be tuned via the amount of monomers and cross-linkers, respectively, and in this way the state of MNP immobilization is influenced. SLP values decreased by up to 40 % in agarose gels for mesh sizes smaller than the hydrodynamic size dH = 20.6 nm. A comparable decrease was observed in polyacrylamide gels (Fig 1b & c). We attribute this drop in SLP values to a gradual immobilization of MNP trapped in the hydrogels, which blocks particle relaxation and therefore decreases heating efficiency. This agrees very well with the results of the in-vitro measurements. The relative difference in the SLP drop in agarose and polyacrylamide hydrogels for similar mesh sizes might be explained by their gel-specific microstructures, which influence the MNP freedom of movement. For validation of these results, further investigations of the relaxation behavior of such trapped MNP via magnetic particle spectroscopy are currently under progress.