https://repositorio.ufu.br/handle/123456789/41072 2026-10-14T18:41:10Z 2026-10-14T18:41:10Z https://repositorio.ufu.br/handle/123456789/50605 2026-10-14T12:50:45Z 2025-12-19T00:00:00Z

Title: Análise da Classificação de Habilidade em Punção Lombar: Uma Abordagem Baseada em Sensores Inerciais e Aprendizado de Máquina Abstract: Lumbar puncture (LP) is an invasive procedure performed to collect cerebrospinal fluid and is commonly used in neurological practice for diagnostic indications and, in some cases, therapeutic purposes. Effective collection depends on the practitioner maintaining a stable hand throughout the procedure; otherwise, complications such as pain, nerve injury, infections, and hemorrhages may occur. The objective of this study is to analyze signals captured by an inertial sensor during the performance of a lumbar puncture procedure on a low-fidelity mannequin and to compare them between participant groups. Method: This study included 26 volunteers, divided into four groups (G1, G2, G3, and G4) composed of medical graduates and senior medical students. Data collection was performed using an inertial sensor (comprising an accelerometer, a gyroscope, and a magnetometer) attached to the participants’ hands, who were wearing surgical gloves. The movement involved simulating the lumbar puncture procedure on a low-fidelity mannequin. Eighteen features were extracted from the collected signals, related to signal amplitude, frequency, entropy, variability, and distribution shape. Group comparisons were conducted using Post-Hoc Kruskal-Wallis tests, and sensitivity, specificity, precision, and accuracy were calculated for the classifiers used: Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbor (KNN), and Naive Bayes (NB). Results: The Post-Hoc Kruskal-Wallis test showed that G3, composed of students without prior LP experience, had the lowest performance, as expected, while G2, composed of physicians who perform LP less frequently, also showed lower performance. Regarding the classifiers, the highest sensitivity was observed in the KNN classifier (95.7%) for the G1/G2 comparison and in the KNN classifier (88.1%) for the G2/G3 comparison; specificity was highest in the KNN classifier (98.4%) for the G2/G4 comparison and in the SVM classifier (78.8%) for the G2/G3 comparison; precision was highest in the KNN classifier (98.8%) and (89%) in the G2/G4 comparison; and accuracy (89%) was highest in the KNN classifier for the G1/G2 comparison.. Conclusion: The use of inertial sensors proved to be a cost-effective alternative to the analysis proposed in this study. The techniques applied for data analysis demonstrated effectiveness in evaluating the collected signals.

2025-12-19T00:00:00Z https://repositorio.ufu.br/handle/123456789/49128 2026-07-29T06:23:46Z 2026-07-10T00:00:00Z

Title: Controle de escorregamento baseado em modelo neuromimético embarcado adaptativo Abstract: The ability to dexterously manipulate objects is a fundamental human skill based on fast reflex loops and tactile feedback. In upper-limb amputees, the lack of sensory perception in commercial prostheses is one of the leading causes of user rejection and device abandonment, with clinical abandonment rates reaching up to 30%. Mimicking the motor grasp reflex in real time (latency below 100 ms) is a major challenge in rehabilitation robotics due to the high computational cost of conventional methods and constraints of wearable hardware. This thesis proposes and validates the hypothesis that an embedded, bio-inspired event-driven architecture based on neuromorphic coding and fed by a high-density piezoresistive tactile array can detect and suppress slip locally and autonomously within timeframes compatible with the human reflex arc. The developed system integrates a 4 × 4 taxel piezoresistive sensor array covered by a 2 mm layer of EcoFlex 00-30 silicone on the fingertips of a commercial robotic prosthesis (COVVI Hand) coupled to an industrial manipulator (DOBOT CR10A). Analog pressure readings are locally converted into action potentials (spikes) using the Izhikevich neuron model to emulate slowly adapting (SA-I) and fast-adapting (FA-I) mechanoreceptors. Active slip suppression is executed using a Monotonic Proportional-Integral (MPI) controller operating at 500 Hz. Under abrupt perturbation (Experiment A: 20 g step load), the average suppression time was 70 ± 22 ms (FA-I) and 80 ± 25 ms (SA-I). Under progressive perturbation (Experiment B: water flow at 50 ml/s), the reaction times were 82 ± 28 ms (FA-I) and 95 ± 34 ms (SA-I). The main finding of this work is the demonstration that local event-driven processing successfully circumvents the non-linearities and hysteresis of the silicone, enabling low-cost microcontrollers to run fast and robust closed-loop reflex controls. In conclusion, the proposed embedded neuromorphic architecture enables energy-efficient wearable solutions, which are essential for real functional upper-limb rehabilitation.

2026-07-10T00:00:00Z https://repositorio.ufu.br/handle/123456789/48733 2026-06-04T06:19:20Z 2025-12-01T00:00:00Z

Title: Efeitos dos Canabinoides no Tremor Parkinsoniano e interações medicamentosas: análise por inteligência artificial explicável dos dados Fox Insight Abstract: Parkinson’s disease constitutes the second most prevalent neurodegenerative condition worldwide, affecting 1-2% of the population over 65 years of age, with tremor present in 70% of cases. Conventional pharmacological treatments present limitations in 30-40% of patients, establishing the need for therapeutic alternatives. This thesis aimed to develop an explainable artificial intelligence methodology for personalizing medical cannabis use in treating parkinsonian tremor, quantifying determinants of individual heterogeneity in therapeutic response. The methodology employed analysis based on SHAP (SHapley Additive exPlanations) values applied to 2,616 records from the Fox Insight platform, structured in three progressive datasets: SINGLE (n=973, isolated cannabis), FULL_NUMERICO (n=931, pharmacological groups), and FULL_GFEM (n=931, detailed individual characteristics). Machine learning models achieved elevated performance metrics: SINGLE (LightGBM) with accuracy 96.31%, precision 96.33%, recall 96.31%, F1-score 96.31%, and specificity 96.03%; FULL_NUMERICO (LightGBM) with accuracy 97.42%, precision 97.47%, recall 97.42%, F1-score 97.42%, and specificity 97.07%; FULL_GFEM (XGBoost) with accuracy 97.42%, precision 97.54%, recall 97.42%, F1-score 97.42%, and specificity 96.91%, demonstrating that interpretability does not compromise predictive performance when architecture is adequately designed. Results quantified eight main findings: (1) pronounced sexual dimorphism, with women presenting mean SHAP values of +1.52 ± 0.48 (improvement rate 94.1%) versus men -0.93 ± 0.37 (rate 38.5%), establishing a magnitude of 2.44x in the probability of favorable response; (2) identification of six distinct profiles through clustering, with 55% of patients presenting elevated improvement rates (89% to 100%) and SHAP values between +1.94 and +2.57; (3) non-linear inverted-U age pattern, with critical group 60-70 years (n=559, 60% of sample) presenting worse response (rate 38.8%) and partial recovery in 70-80 years (rate 76.2%, n=181); (4) hierarchy of administration routes quantified by MASV (Mean Absolute SHAP Value): oil (0.705) > sublingual (0.4959) > food (0.657 Class 1), demonstrating importance superior to dose; (5) bifasic optimal therapeutic windows for THC (1.5-2.5 mg/day, MASV = 0.2888) and CBD (2-4 mg/day, MASV = 0.2654), empirically validating that supra-therapeutic doses reduce efficacy; (6) quantified synergy with MAO-B inhibitors (MASV = 0.4485), superior to interactions with levodopa (MASV = 0.2506) and other dopaminergics (MASV = 0.2454); (7) moderate optimal frequency of 3-5 times/week, superior to continuous daily use; (8) therapeutic learning curve with optimization in 2-3 years and possible tolerance after >3 years. Findings were operationalized in a clinical nomogram that stratifies patients into five predictive categories, with 87.3% concordance with phenotypic clusters. Computational analysis through clustering identified difference in improvement rates between subgroups with elevated response (89-100%, n=513) and general population (66.4%, n=931), suggesting potential for therapeutic personalization based on identifiable clinical characteristics, hypothesis requiring confirmation in future clinical studies. This investigation establishes a new methodological standard for research in precision neurological medicine, providing computational tools and quantitative knowledge for prospective identification of subgroups with high probability of benefit, individualized therapeutic regimen optimization, and establishment of expectations based on measurable patient characteristics.

2025-12-01T00:00:00Z https://repositorio.ufu.br/handle/123456789/48707 2026-05-20T06:29:41Z 2026-05-04T00:00:00Z

Title: Avaliação da atividade eletromiográfica de músculos do core no exercício prancha ventral até a exaustão com e sem exergame plankpad Abstract: Core training plays an important role in quality of life, especially in reducing the prevalence of low back pain. Recently, exergames such as Plankpad have been developed to improve adherence to physical training and rehabilitation programs focused on core muscle strengthening. However, no studies have analyzed electromyographic (EMG) activity during the ventral plank (VP) exercise with and without exergame, including time-domain and frequency-domain analysis as well as muscle co-contraction. This study aimed to investigate the effects of incorporating the Plankpad exergame on core muscle EMG activity during the VP exercise until exhaustion. Fifteen young, healthy, physically active men with no history of low back pain participated. Participants performed the VP exercise until exhaustion under five conditions: stable ground (VP-GR), Plankpad platform without game (VP-PP), and Plankpad platform with the Fruit Splicer game at easy (VP-EA), medium (VP-ME), and hard (VP-HA) levels. EMG signals were collected from the rectus abdominis (RA), external oblique (EO), internal oblique (IO), erector spinae (ES), and multifidus (MU) muscles, along with time to exhaustion and rating of perceived exertion (RPE). Results showed progressive increases in RMS and decreases in median frequency (MF) over time in all exercises, indicating peripheral neuromuscular fatigue, with no significant differences among the five conditions for most parameters. Local co-contraction (IO/MU) was significantly higher in VP-PP compared to VP EA and decreased with fatigue across all exercises. Time to exhaustion and RPE were similar among conditions. Correlations between RPE and EMG parameters were weak to moderate. We conclude that the Plankpad exergame, at easy, medium, and hard levels, does not increase electromyographic activation, spectral fatigue, time to exhaustion, or RPE compared to traditional plank exercise in young healthy trained men.

2026-05-04T00:00:00Z