Search

The integration of deep learning into radiology has the potential to enhance diagnostic processes, yet its acceptance in clinical practice remains limited due to various challenges. This study aimed to develop and evaluate a fine-tuned large language model (LLM), based on Llama 3-8B, to automate the generation of accurate and concise conclusions in magnetic resonance imaging (MRI) and computed tomography (CT) radiology reports, thereby assisting radiologists and improving reporting efficiency. A dataset comprising 15,000 radiology reports was collected from the University of Medicine and Pharmacy of Craiova’s Imaging Center, covering a diverse range of MRI and CT examinations made by four experienced radiologists. The Llama 3-8B model was fine-tuned using transfer-learning techniques, incorporating parameter quantization to 4-bit precision and low-rank adaptation (LoRA) with a rank of 16 to optimize computational efficiency on consumer-grade GPUs. The model was trained over five epochs using an NVIDIA RTX 3090 GPU, with intermediary checkpoints saved for monitoring. Performance was evaluated quantitatively using Bidirectional Encoder Representations from Transformers Score (BERTScore), Recall-Oriented Understudy for Gisting Evaluation (ROUGE), Bilingual Evaluation Understudy (BLEU), and Metric for Evaluation of Translation with Explicit Ordering (METEOR) metrics on a held-out test set. Additionally, a qualitative assessment was conducted, involving 13 independent radiologists who participated in a Turing-like test and provided ratings for the AI-generated conclusions. The fine-tuned model demonstrated strong quantitative performance, achieving a BERTScore F1 of 0.8054, a ROUGE-1 F1 of 0.4998, a ROUGE-L F1 of 0.4628, and a METEOR score of 0.4282. In the human evaluation, the artificial intelligence (AI)-generated conclusions were preferred over human-written ones in approximately 21.8% of cases, indicating that the model’s outputs were competitive with those of experienced radiologists. The average rating of the AI-generated conclusions was 3.65 out of 5, reflecting a generally favorable assessment. Notably, the model maintained its consistency across various types of reports and demonstrated the ability to generalize to unseen data. The fine-tuned Llama 3-8B model effectively generates accurate and coherent conclusions for MRI and CT radiology reports. By automating the conclusion-writing process, this approach can assist radiologists in reducing their workload and enhancing report consistency, potentially addressing some barriers to the adoption of deep learning in clinical practice. The positive evaluations from independent radiologists underscore the model’s potential utility. While the model demonstrated strong performance, limitations such as dataset bias, limited sample diversity, a lack of clinical judgment, and the need for large computational resources require further refinement and real-world validation. Future work should explore the integration of such models into clinical workflows, address ethical and legal considerations, and extend this approach to generate complete radiology reports.

This article starts from the premise that one of the global control strategies of the Permanent Magnet Synchronous Motor (PMSM), namely the Direct Torque Control (DTC) control strategy, is characterized by the fact that the internal flux and torque control loop usually uses ON–OFF controllers with hysteresis, which offer easy implementation and very short response times, but the oscillations introduced by them must be cancelled by the external speed loop controller. Typically, this is a PI speed controller, whose performance is good around global operating points and for relatively small variations in external parameters and disturbances, caused in particular by load torque variation. Exploiting the advantages of the DTC strategy, this article presents a way to improve the performance of the sensorless control system (SCS) of the PMSM using the Proportional Integrator (PI), PI Equilibrium Optimizer Algorithm (EOA), Fractional Order (FO) PI, Tilt Integral Derivative (TID) and FO Lead–Lag under constant flux conditions. Sliding Mode Control (SMC) and FOSMC are proposed under conditions where the flux is variable. The performance indicators of the control system are the usual ones: response time, settling time, overshoot, steady-state error and speed ripple, plus another one given by the fractal dimension (FD) of the PMSM rotor speed signal, and the hypothesis that the FD of the controlled signal is higher when the control system performs better is verified. The article also presents the basic equations of the PMSM, based on which the synthesis of integer and fractional controllers, the synthesis of an observer for estimating the PMSM rotor speed, electromagnetic torque and stator flux are presented. The comparison of the performance for the proposed control systems and the demonstration of the parametric robustness are performed by numerical simulations in Matlab/Simulink using Simscape Electrical and Fractional-Order Modelling and Control (FOMCON). Real-time control based on an embedded system using a TMS320F28379D controller demonstrates the good performance of the PMSM-SCS based on the DTC strategy in a complete Hardware-In-the-Loop (HIL) implementation.

Currently, medical imaging has largely supplanted traditional methods in the realm of diagnosis and treatment planning. This shift is primarily attributable to the non-invasive nature, rapidity, and user-friendliness of medical-imaging techniques. The widespread adoption of medical imaging, however, has shifted the bottleneck to healthcare professionals who must analyze each case post-image acquisition. This process is characterized by its sluggishness and subjectivity, making it susceptible to errors. The anterior cruciate ligament (ACL), a frequently injured knee ligament, predominantly affects a youthful and sports-active demographic. ACL injuries often leave patients with substantial disabilities and alter knee mechanics. Since some of these cases necessitate surgery, it is crucial to accurately classify and detect ACL injury. This paper investigates the utilization of pre-trained convolutional neural networks featuring residual connections (ResNet) along with image-processing methods to identify ACL injury and differentiate between various tear levels. The ResNet employed in this study is not the standard ResNet but rather an adapted version capable of processing 3D volumes constructed from 2D image slices. Achieving a peak accuracy of 97.15% with a custom split, 96.32% through Monte-Carlo cross-validation, and 93.22% via five-fold cross-validation, our approach enhances the performance of three-class classifiers by over 7% in terms of raw accuracy. Moreover, we achieved an improvement of more than 1% across all types of evaluation. It is quite clear that the model’s output can effectively serve as an initial diagnostic baseline for radiologists with minimal effort and nearly instantaneous results. This advancement underscores the paper’s focus on harnessing deep learning for the nuanced detection and classification of ACL tears, demonstrating a significant leap toward automating and refining diagnostic accuracy in sports medicine and orthopedics.

Starting from the problem of studying the parametric robustness in the case of the control of a permanent magnet-synchronous motor (PMSM), although robust control systems correspond entirely to this problem, due to the complexity of the algorithms of the robust type, in this article the use of switched systems theory is proposed as a study option, given the fact that these types of systems are suitable both for the study of systems with variable structure and for systems with significant parametric variation under conditions of lower complexity of the control algorithms. The study begins by linearizing a PMSM model at a static operating point and continues with a systematic presentation of the basic elements and concepts concerning the stability of switched systems by applying these concepts to the control system of a PMSM based on the field-oriented control (FOC) strategy, which usually changes the value of its parameters during operation (stator resistance Rs, stator inductances Ld and Lq, but also combined inertia of PMSM rotor and load J). The numerical simulations performed in Simulink validate the fact that, for parametric variations of the PMSM structure, the PMSM control switched systems preserve qualitative performance in terms of its control. A series of Matlab programs are presented based on the YALMIP toolbox to obtain Pi matrices, by solving Lyapunov–Metzler type inequalities, and using dwell time to demonstrate stability, as well as the qualitative study of the performance of PMSM control switched systems by presenting in phase plane and state space analysis of the evolution of state vectors: ω PMSM rotor speed, iq current, and id current.

The vehicle suspension system is represented by a complex group of components which connect the wheels to the frame or body. Its primary function is to reduce or absorb various vehicle vibrations generated by road disturbances, providing comfort and safety for passengers. Most modern vehicles have independent, active, or semiactive front and rear suspensions which allow the use of electronic actuation. For this reason, automotive engineers conduct research on the active suspension model to determine the most suitable control algorithm. Three active suspension models are intensely used within simulations: the quarter-car, the half-car, and the full-car models. This paper proposes an adaptive harmonic control for a half-car active suspension system. The mathematical model of the suspension system is implemented in the MATLAB/Simulink environment. The control approach is tested via simulation, and several comparisons with the classical proportional-integral controller are provided. The simulation results show that the controller behaves quite similarly on the half-car model as it did on a quarter-car model. Additionally, an improvement to the harmonic control algorithm has been accomplished.

Remote monitoring and operation evaluation applications for industrial environments are modern and easy means of exploiting the provided resources of specific systems. Targeted micro hydropower plant functionalities (such as tracking and adjusting the values of functional parameters, real-time fault and cause signalizing, condition monitoring assurance, and assessments of the need for maintenance activities) require the design of reliable and efficient devices or systems. The present work describes the design and implementation procedure of an Industrial Internet of Things (IIoT) system configured for a basic micro hydropower plant architecture and assuring simple means of customization for plant differences in structure and operation. The designed system features a set of commonly used functions specific to micro hydropower exploitation, providing maximum performance and efficiency.

Image processing remains an area that has impact on the software industry and is a field that is permanently developing in both IT and industrial contexts. Nowadays, the demand for fast computing times is becoming increasingly difficult to fulfill in the case of massive computing systems. This article proposes a particular case of efficiency for a specifically developed model for fractal generations. From the point of view of graphic analysis, the application can generate a series of fractal images. This process is analyzed and compared in this study from a programming perspective in terms of both the results at the processor level and the graphical generation possibilities. This paper presents the structure of the software and its implementation for generating fractal images using the Mandelbrot set. Starting from the complex mathematical set, the component iterations of the Mandelbrot algorithm lead to optimization variants for the calculation. The article consists of a presentation of an optimization variant based on applying parallel calculations for fractal generation. The method used in the study assumes a high grade of accuracy regarding the selected mathematical model for fractal generation and does not characterize a method specially built for a certain kind of image. A series of scenarios are analyzed, and details related to differences in terms of calculation times, starting from the more efficient proposed variant, are presented. The developed software implementation is parallelization-based and is optimized for generating a wide variety of fractal images while also providing a test package for the generated environment. The influence of parallel programming is highlighted in terms of its difference to sequential programming to, in turn, highlight recent methods of speeding up computing times. The purpose of the article is to combine the complexity of the mathematical calculation behind the fractal sets with programming techniques to provides an analysis of the graphic results from the point of view of the use of computing resources and working time.

The specific equipment, installation and machinery infrastructure of an electric power system have always required specially designed data acquisition systems and devices to ensure their safe operation and monitoring. Besides maintenance, periodical upgrade must be ensured for these systems, to meet the current practical requirements. Monitoring, testing, and diagnosis altogether represent key activities in the development process of electric power elements. This work presents the detailed structure and implementation of a complex, configurable system which can assure efficient monitoring, testing, and diagnosis for various electric power infrastructures, with proven efficiency through a comprehensive set of experimental results obtained in real running conditions. The developed hardware and software implementation is a robust structure, optimized for acquiring a large variety of electrical signals, also providing easy and fast connection within the monitored environment. Its high level of configurability and very good price–performance ratio makes it an original and handy solution for electric power infrastructures.

Background and Objectives: Malignant bone tumors represent a major problem due to their aggressiveness and low survival rate. One of the determining factors for improving vital and functional prognosis is the shortening of the time between the onset of symptoms and the moment when treatment starts. The objective of the study is to predict the malignancy of a bone tumor from magnetic resonance imaging (MRI) using deep learning algorithms. Materials and Methods: The cohort contained 23 patients in the study (14 women and 9 men with ages between 15 and 80). Two pretrained ResNet50 image classifiers are used to classify T1 and T2 weighted MRI scans. To predict the malignancy of a tumor, a clinical model is used. The model is a feed forward neural network whose inputs are patient clinical data and the output values of T1 and T2 classifiers. Results: For the training step, the accuracies of 93.67% for the T1 classifier and 86.67% for the T2 classifier were obtained. In validation, both classifiers obtained 95.00% accuracy. The clinical model had an accuracy of 80.84% for training phase and 80.56% for validation. The receiver operating characteristic curve (ROC) of the clinical model shows that the algorithm can perform class separation. Conclusions: The proposed method is based on pretrained deep learning classifiers which do not require a manual segmentation of the MRI images. These algorithms can be used to predict the malignancy of a tumor and on the other hand can shorten the time of their diagnosis and treatment process. While the proposed method requires minimal intervention from an imagist, it needs to be tested on a larger cohort of patients.

The field-oriented control (FOC) strategy of a permanent magnet synchronous motor (PMSM) in a simplified form is based on PI-type controllers. In addition to their low complexity (an advantage for real-time implementation), these controllers also provide limited performance due to the nonlinear character of the description equations of the PMSM model under the usual conditions of a relatively wide variation in the load torque and the high dynamics of the PMSM speed reference. Moreover, a number of significant improvements in the performance of PMSM control systems, also based on the FOC control strategy, are obtained if the controller of the speed control loop uses sliding mode control (SMC), and if the controllers for the inner control loops of id and iq currents are of the synergetic type. Furthermore, using such a control structure, very good performance of the PMSM control system is also obtained under conditions of parametric uncertainties and significant variations in the combined rotor-load moment of inertia and the load resistance. To improve the performance of the PMSM control system without using controllers having a more complicated mathematical description, the advantages provided by reinforcement learning (RL) for process control can also be used. This technique does not require the exact knowledge of the mathematical model of the controlled system or the type of uncertainties. The improvement in the performance of the PMSM control system based on the FOC-type strategy, both when using simple PI-type controllers or in the case of complex SMC or synergetic-type controllers, is achieved using the RL based on the Deep Deterministic Policy Gradient (DDPG). This improvement is obtained by using the correction signals provided by a trained reinforcement learning agent, which is added to the control signals ud, uq, and iqref. A speed observer is also implemented for estimating the PMSM rotor speed. The PMSM control structures are presented using the FOC-type strategy, both in the case of simple PI-type controllers and complex SMC or synergetic-type controllers, and numerical simulations performed in the MATLAB/Simulink environment show the improvements in the performance of the PMSM control system, even under conditions of parametric uncertainties, by using the RL-DDPG.

This article presents the control of a grid connected PV (GC-PV) array system, starting from a benchmark. The control structure used in this article was a cascade-type structure, in which PI or synergetic (SYN) controllers were used for the inner control loop of id and iq currents and PI or sliding mode control (SMC) controllers were used for the outer control loop of the udc voltage from the DC intermediate circuit. This paper presents the mathematical model of the PV array together with the main component blocks: simulated inputs for the PV array; the PV array itself; the MPPT algorithm; the DC-DC boost converter; the voltage and current measurements for the DC intermediate circuit; the load and connection to power grid; the DC-AC converter; and the power grid. It also presents the stages of building and training the reinforcement learning (RL) agent. To improve the performance of the control system for the GC-PV array system without using controllers with a more complicated mathematical description, the advantages provided by the RL agent on process controls could also be used. This technique does not require exact knowledge of the mathematical model of the controlled system or the type of uncertainties. The improvement in the control system performance for the GC-PV array system, both when using simple PI-type controllers or complex SMC- and SYN-type controllers, was achieved using an RL agent based on the Deep Deterministic Policy Gradient (DDPG). The variant of DDPG used in this study was the Twin-Delayed (TD3). The improvement in performance of the control system were obtained by using the correction command signals provided by the trained RL agent, which were added to the command signals ud, uq and idref. The parametric robustness of the proposed control system based on SMC and SYN controllers for the GC-PV array system was proven in the case of a variation of 30% caused by the three-phase load. Moreover, the results of the numerical simulations are shown comparatively and the validation of the synthesis of the proposed control system was obtained. This was achieved by comparing the proposed system with a software benchmark for the control of a GC-PV array system performed in MATLAB Simulink. The numerical simulations proved the superiority of the performance of control systems that use the RL-TD3 agent.

This work proposes a realistic solution to the control problem of sensorless induction motors. Due to some important aspects related to their construction and reliability, the induction motors are extensively used in many modern industrial applications. Considering that the system is facing the lack of hardware sensors, the proposed complex control strategies are based on the estimation of unavailable system variables and parameters. In order to control the rotor speed, two robust control strategies are proposed: a modified super-twisting adaptive technique and a model predictive technique. The tests performed under several practical assumptions show that the closed loop behaviour of the system is adequate, and the output variable follows the imposed time varying reference, despite the considered uncertainties and disturbances acting on the process.

This work addresses the analysis and design of various Proportional-Integral-Derivative (PID) control techniques for a three-level inverter. Multilevel power converters are modern and basic elements of high-voltage electric drive and power supply systems. By using simulations and specific computer-aided design tools, the overall functional characteristics of multilevel converters, as well as the electrical demands of the components, can be accurately assessed to obtain an appropriate control solution. An innovative and detailed software model of a three-level inverter is developed and then used for the implementation of control techniques. Several tuning methods are used to tune PID controllers for two specific cases: the multilevel inverter with a linear load and with an asynchronous motor load, respectively. A detailed analysis and comparisons of the quality criteria and control performance are achieved. This analysis shows that the choice of controller type depends on the inverter load. For the linear load, proper results are obtained with a PI Nichols-tuned controller, and for the asynchronous load, with a PI controller tuned via a modified Hokushin method. The computer-aided design tools can be further used for the simulation of the equipment in various operating conditions, normal and fault, following all functional parameters.

This paper presents the design of a nonlinear multivariable adaptive control strategy for a class of wastewater depollutionbioprocesses that are carried out in recycle bioreactors. The adaptive control structure is achieved by combining a linearizing controllaw with a new state observer and a parameter estimator, which play the role of the software sensors for on-line estimation of biologicalstates and parameter variables of interest of the bioprocess. Simulation results are included to illustrate the performance of bothestimation and control algorithms.

This paper presents the control problem of a class of propagation bioprocesses that are carried out in fixed bed reactors. Since the dynamics of these processes are described by partial differential equations, in order to obtain useful models for control purposes, a possible method consists of approximation of their infinitely order associated models by finite order models. A class of nonlinear adaptive controllers are then designed based on these finite order models, which consist of a set of ordinary differential equations obtained here by orthogonal collocation method. Computer simulations conducted in the case of a fixed bed reactor are included to illustrate the performances of the proposed adaptive controllers.

The paper presents a robust control design strategy for bioprocesses, which are characterized by strongly nonlinear dynamics. More precisely, we present the H2 methodology in order to compute the controller for a recycle Continuous Stirred Tank Bioreactor (CSTB). We consider a general method of formulating control problem, which makes use of linear fractional transformation as introduced by Doyle (1978). The formulation makes use of the general two-port configuration of the generalized plant with a generalized controller. The H2 norm is the quadratic criterion used in optimal control as LQG. The overall control objective is to minimize the H2 norm of the transfer matrix function from the weighted exogenous inputs to the weighted controlled outputs. The advantage of H2 control technique, which uses the linearized model of the CSTB, is that it is completely automated and very flexible. Finally, we prove that the closed loop control structure has very good inner robustness.

The paper presents on-line estimation strategies for bioprocesses, which are characterized by strongly nonlinear dynamics. A general form for the state observers is analysed and the exponential observability of bioprocesses is discussed. For those biotechnological processes that possess the property of exponential observability, an extended Luenberger observer is implemented. If exponential observers cannot be used, a solution is the design of asymptotic observers. The proposed observers are implemented for a microbial growth process coupled with an enzyme-catalysed reaction, which is a usual bioprocess that takes place in a fed-batch bioreactor. Illustrative computer simulations and conclusions are included.