RESUMO
AIM: To test the feasibility and accuracy of a new attention-based deep learning (DL) method for right ventricular (RV) quantification using 2D echocardiography (2DE) with cardiac magnetic resonance imaging (CMR) as reference. METHODS AND RESULTS: We retrospectively analyzed images from 50 adult patients (median age 51, interquartile range 32-62 42% women) who had undergone CMR within 1 month of 2DE. RV planimetry of the myocardial border was performed in end-diastole (ED) and end-systole (ES) for eight standardized 2DE RV views with calculation of areas. The DL model comprised a Feature Tokenizer module and a stack of Transformer layers. Age, gender and calculated areas were used as inputs, and the output was RV volume in ED/ES. The dataset was randomly split into training, validation and testing subsets (35, 5 and 10 patients respectively). Mean RVEDV, RVESV and RV ejection fraction (EF) were 163 ± 70 mL, 82 ± 42 mL and 51% ± 8% respectively without differences among the subsets. The proposed method achieved good prediction of RV volumes (R2 = .953, absolute percentage error [APE] = 9.75% ± 6.23%) and RVEF (APE = 7.24% ± 4.55%). Per CMR, there was one patient with RV dilatation and three with RV dysfunction in the testing dataset. The DL model detected RV dilatation in 1/1 case and RV dysfunction in 4/3 cases. CONCLUSIONS: An attention-based DL method for 2DE RV quantification showed feasibility and promising accuracy. The method requires validation in larger cohorts with wider range of RV size and function. Further research will focus on the reduction of the number of required 2DE to make the method clinically applicable.
Assuntos
Aprendizado Profundo , Disfunção Ventricular Direita , Adulto , Feminino , Humanos , Masculino , Pessoa de Meia-Idade , Ecocardiografia , Estudos de Viabilidade , Reprodutibilidade dos Testes , Estudos Retrospectivos , Volume Sistólico , Função Ventricular DireitaRESUMO
BACKGROUND: Machine learning (ML) is increasingly being applied in Cardiology to predict outcomes and assist in clinical decision-making. We sought to develop and validate an ML model for the prediction of mortality after heart transplantation (HT) in adults with congenital heart disease (ACHD). METHODS: The United Network for Organ Sharing (UNOS) database was queried from 2000 to 2020 for ACHD patients who underwent isolated HT. The study cohort was randomly split into derivation (70%) and validation (30%) datasets that were used to train and test a CatBoost ML model. Feature selection was performed using SHapley Additive exPlanations (SHAP). Recipient, donor, procedural, and post-transplant characteristics were tested for their ability to predict mortality. We additionally used SHAP for explainability analysis, as well as individualized mortality risk assessment. RESULTS: The study cohort included 1033 recipients (median age 34 years, 61% male). At 1 year after HT, there were 205 deaths (19.9%). Out of a total of 49 variables, 10 were selected as highly predictive of 1-year mortality and were used to train the ML model. Area under the curve (AUC) and predictive accuracy for the 1-year ML model were .80 and 75.2%, respectively, and .69 and 74.2% for the 3-year model, respectively. Based on SHAP analysis, hemodialysis of the recipient post-HT had overall the strongest relative impact on 1-year mortality after HΤ, followed by recipient-estimated glomerular filtration rate, age and ischemic time. CONCLUSIONS: ML models showed satisfactory predictive accuracy of mortality after HT in ACHD and allowed for individualized mortality risk assessment.
Assuntos
Cardiopatias Congênitas , Insuficiência Cardíaca , Transplante de Coração , Humanos , Masculino , Adulto , Feminino , Medição de Risco , Cardiopatias Congênitas/cirurgia , Aprendizado de MáquinaRESUMO
BACKGROUND: Recent studies demonstrated safety and efficacy of heart transplantation (HT) from hepatitis C virus (HCV)-positive donors. We sought to evaluate the impact of HCV donor status on the outcomes of patients undergoing HT in the United States. METHODS: We analyzed a retrospective cohort of adult patients from the United Network for Organ Sharing (UNOS) database who underwent isolated HT from 2015 until present. Primary outcomes were 30-day and 1-year overall mortality. Secondary outcomes included risk for graft failure and overall survival, incident stroke and need for dialysis during the available follow-up period. All end points were evaluated according to HCV status. RESULTS: All-cause 30-day and 1-year mortality was similar between the two groups (3.4% vs 3.2%, P = .973 and 6.9% vs 7.8%, P = .769, respectively, for patients receiving heart grafts from HCV+ vs. HCV- donors). Graft failure was 12.8% (95% CI: 8%-19%) and 15.2% (95 CI: 15%-16%) in the HCV+ and HCV- groups, respectively (P = .92 and P = .68). Competing risk regression analysis for re-operation showed a non-significant trend for higher risk for re-transplantation in the HCV+ group (HR: 2.71; 95% CI: 0.83, 8.80, P = .097). CONCLUSION: HCV donor status does not seem to negatively affect the outcomes of HT in the U.S population.
Assuntos
Transplante de Coração , Hepatite C , Adulto , Sobrevivência de Enxerto , Hepatite C/etiologia , Humanos , Estudos Retrospectivos , Doadores de Tecidos , Resultado do Tratamento , Estados Unidos/epidemiologiaRESUMO
PURPOSE: We sought to develop and validate machine learning (ML) models to increase the predictive accuracy of mortality after heart transplantation (HT). METHODS AND RESULTS: We included adult HT recipients from the United Network for Organ Sharing (UNOS) database between 2010 and 2018 using solely pre-transplant variables. The study cohort comprised 18 625 patients (53 ± 13 years, 73% males) and was randomly split into a derivation and a validation cohort with a 3:1 ratio. At 1-year after HT, there were 2334 (12.5%) deaths. Out of a total of 134 pre-transplant variables, 39 were selected as highly predictive of 1-year mortality via feature selection algorithm and were used to train five ML models. AUC for the prediction of 1-year survival was .689, .642, .649, .637, .526 for the Adaboost, Logistic Regression, Decision Tree, Support Vector Machine, and K-nearest neighbor models, respectively, whereas the Index for Mortality Prediction after Cardiac Transplantation (IMPACT) score had an AUC of .569. Local interpretable model-agnostic explanations (LIME) analysis was used in the best performing model to identify the relative impact of key predictors. ML models for 3- and 5-year survival as well as acute rejection were also developed in a secondary analysis and yielded AUCs of .629, .609, and .610 using 27, 31, and 91 selected variables respectively. CONCLUSION: Machine learning models showed good predictive accuracy of outcomes after heart transplantation.
Assuntos
Transplante de Coração , Aprendizado de Máquina , Adulto , Idoso , Algoritmos , Área Sob a Curva , Bases de Dados Factuais , Feminino , Humanos , Masculino , Pessoa de Meia-IdadeRESUMO
Global competition among businesses imposes a more effective and low-cost supply chain allowing firms to provide products at a desired quality, quantity, and time, with lower production costs. The latter include holding cost, ordering cost, and backorder cost. Backorder occurs when a product is temporarily unavailable or out of stock and the customer places an order for future production and shipment. Therefore, stock unavailability and prolonged delays in product delivery will lead to additional production costs and unsatisfied customers, respectively. Thus, it is of high importance to develop models that will effectively predict the backorder rate in an inventory system with the aim of improving the effectiveness of the supply chain and, consequentially, the performance of the company. However, traditional approaches in the literature are based on stochastic approximation, without incorporating information from historical data. To this end, machine learning models should be employed for extracting knowledge of large historical data to develop predictive models. Therefore, to cover this need, in this study, the backorder prediction problem was addressed. Specifically, various machine learning models were compared for solving the binary classification problem of backorder prediction, followed by model calibration and a post-hoc explainability based on the SHAP model to identify and interpret the most important features that contribute to material backorder. The results showed that the RF, XGB, LGBM, and BB models reached an AUC score of 0.95, while the best-performing model was the LGBM model after calibration with the Isotonic Regression method. The explainability analysis showed that the inventory stock of a product, the volume of products that can be delivered, the imminent demand (sales), and the accurate prediction of the future demand can significantly contribute to the correct prediction of backorders.
Assuntos
Aprendizado de MáquinaRESUMO
In this paper, we propose a dense multi-scale adaptive graph convolutional network (DMA-GCN) method for automatic segmentation of the knee joint cartilage from MR images. Under the multi-atlas setting, the suggested approach exhibits several novelties, as described in the following. First, our models integrate both local-level and global-level learning simultaneously. The local learning task aggregates spatial contextual information from aligned spatial neighborhoods of nodes, at multiple scales, while global learning explores pairwise affinities between nodes, located globally at different positions in the image. We propose two different structures of building models, whereby the local and global convolutional units are combined by following an alternating or a sequential manner. Secondly, based on the previous models, we develop the DMA-GCN network, by utilizing a densely connected architecture with residual skip connections. This is a deeper GCN structure, expanded over different block layers, thus being capable of providing more expressive node feature representations. Third, all units pertaining to the overall network are equipped with their individual adaptive graph learning mechanism, which allows the graph structures to be automatically learned during training. The proposed cartilage segmentation method is evaluated on the entire publicly available Osteoarthritis Initiative (OAI) cohort. To this end, we have devised a thorough experimental setup, with the goal of investigating the effect of several factors of our approach on the classification rates. Furthermore, we present exhaustive comparative results, considering traditional existing methods, six deep learning segmentation methods, and seven graph-based convolution methods, including the currently most representative models from this field. The obtained results demonstrate that the DMA-GCN outperforms all competing methods across all evaluation measures, providing DSC=95.71% and DSC=94.02% for the segmentation of femoral and tibial cartilage, respectively.
RESUMO
Atrial fibrillation (AF) significantly increases the risk of stroke and heart failure, but is frequently asymptomatic and intermittent; therefore, its timely diagnosis poses challenges. Early detection in selected patients may aid in stroke prevention and mitigate structural heart complications through prompt intervention. Smartwatches, coupled with powerful artificial intelligence (AI)-enabled algorithms, offer a promising tool for early detection due to their widespread use, easiness of use, and potential cost-effectiveness. Commercially available smartwatches have gained clearance from the FDA to detect AF and are becoming increasingly popular. Despite their promise, the evolving landscape of AI-enabled smartwatch-based AF detection raises questions about the clinical value of this technology. Following the ongoing digital transformation of healthcare, clinicians should familiarize themselves with how AI-enabled smartwatches function in AF detection and navigate their role in clinical settings to deliver optimal patient care. In this review, we provide a concise overview of the characteristics of AI-enabled smartwatch algorithms, their diagnostic performance, clinical value, limitations, and discuss future perspectives in AF diagnosis.
RESUMO
The main goal driving this work is to develop computer-aided classification models relying on clinical data to identify coronary artery disease (CAD) instances with high accuracy while incorporating the expert's opinion as input, making it a "man-in-the-loop" approach. CAD is traditionally diagnosed in a definite manner by Invasive Coronary Angiography (ICA). A dataset was created using biometric and clinical data from 571 patients (21 total features, 43% ICA-confirmed CAD instances) along with the expert's diagnostic yield. Five machine learning classification algorithms were applied to the dataset. For the selection of the best feature set for each algorithm, three different parameter selection algorithms were used. Each ML model's performance was evaluated using common metrics, and the best resulting feature set for each is presented. A stratified ten-fold validation was used for the performance evaluation. This procedure was run both using the assessments of experts/doctors as input and without them. The significance of this paper lies in its innovative approach of incorporating the expert's opinion as input in the classification process, making it a "man-in-the-loop" approach. This approach not only increases the accuracy of the models but also provides an added layer of explainability and transparency, allowing for greater trust and confidence in the results. Maximum achievable accuracy, sensitivity, and specificity are 83.02%, 90.32%, and 85.49% when using the expert's diagnosis as input, compared to 78.29%, 76.61%, and 86.07% without the expert's diagnosis. The results of this study demonstrate the potential for this approach to improve the diagnosis of CAD and highlight the importance of considering the role of human expertise in the development of computer-aided classification models.
Assuntos
Doença da Artéria Coronariana , Humanos , Doença da Artéria Coronariana/diagnóstico , Algoritmos , Angiografia Coronária , Aprendizado de Máquina , BiometriaRESUMO
Despite therapeutic advancements, stroke remains a leading cause of death and long-term disability. The quality of current stroke prognostic models varies considerably, whereas prediction models of post-stroke disability and mortality are restricted by the sample size, the range of clinical and risk factors and the clinical applicability in general. Accurate prognostication can ease post-stroke discharge planning and help healthcare practitioners individualize aggressive treatment or palliative care, based on projected life expectancy and clinical course. In this study, we aimed to develop an explainable machine learning methodology to predict functional outcomes of stroke patients at discharge, using the Modified Rankin Scale (mRS) as a binary classification problem. We identified 35 parameters from the admission, the first 72 h, as well as the medical history of stroke patients, and used them to train the model. We divided the patients into two classes in two approaches: "Independent" vs. "Non-Independent" and "Non-Disability" vs. "Disability". Using various classifiers, we found that the best models in both approaches had an upward trend, with respect to the selected biomarkers, and achieved a maximum accuracy of 88.57% and 89.29%, respectively. The common features in both approaches included: age, hemispheric stroke localization, stroke localization based on blood supply, development of respiratory infection, National Institutes of Health Stroke Scale (NIHSS) upon admission and systolic blood pressure levels upon admission. Intubation and C-reactive protein (CRP) levels upon admission are additional features for the first approach and Erythrocyte Sedimentation Rate (ESR) levels upon admission for the second. Our results suggest that the said factors may be important predictors of functional outcomes in stroke patients.
RESUMO
In the last few years, deep learning has made a breakthrough and established its position in machine learning classification problems in medical image analysis. Deep learning has recently displayed remarkable applicability in a range of different medical applications, as well as in nuclear cardiology. This paper implements a literature review protocol and reports the latest advances in artificial intelligence (AI)-based classification in SPECT myocardial perfusion imaging in heart disease diagnosis. The representative and most recent works are reported to demonstrate the use of AI and deep learning technologies in medical image analysis in nuclear cardiology for cardiovascular diagnosis. This review also analyses the primary outcomes of the presented research studies and suggests future directions focusing on the explainability of the deployed deep-learning systems in clinical practice.
Assuntos
Aprendizado Profundo , Imagem de Perfusão do Miocárdio , Inteligência Artificial , Algoritmos , Tomografia Computadorizada de Emissão de Fóton ÚnicoRESUMO
Deep learning (DL) has a growing popularity and is a well-established method of artificial intelligence for data processing, especially for images and videos. Its applications in nuclear medicine are broad and include, among others, disease classification, image reconstruction, and image de-noising. Positron emission tomography (PET) and single-photon emission computerized tomography (SPECT) are major image acquisition technologies in nuclear medicine. Though several studies have been conducted to apply DL in many nuclear medicine domains, such as cancer detection and classification, few studies have employed such methods for cardiovascular disease applications. The present paper reviews recent DL approaches focused on cardiac SPECT imaging. Extensive research identified fifty-five related studies, which are discussed. The review distinguishes between major application domains, including cardiovascular disease diagnosis, SPECT attenuation correction, image denoising, full-count image estimation, and image reconstruction. In addition, major findings and dominant techniques employed for the mentioned task are revealed. Current limitations of DL approaches and future research directions are discussed.
RESUMO
Identifying and measuring soccer playing styles is a very important step toward a more effective performance analysis. Exploring the different game styles that a team can adopt to enable a great performance remains under-researched. To address this challenge and identify new directions in future research in the area, this paper conducted a critical review of 40 research articles that met specific criteria. Following the 22-item Preferred Reporting Items for Systematic Reviews and Meta-Analyses extension for Scoping Reviews (PRISMA-ScR) guidelines, this scoping review searched for literature on Google Scholar and Pub Med database. The descriptive and thematic analysis found that the objectives of the identified papers can be classified into three main categories (recognition and effectiveness of playing styles and contextual variables that affect them). Critically reviewing the studies, the paper concluded that: (i) factor analysis seems to be the best technique among inductive statistics; (ii) artificial intelligence (AI) opens new horizons in performance analysis, and (iii) there is a need for further research on the effectiveness of different playing styles, as well as on the impact of contextual variables on them.
RESUMO
Identifying playing styles in football is highly valuable for achieving effective performance analysis. While there is extensive research on team styles, studies on individual player styles are still in their early stages. Thus, the aim of this systematic review was to provide a comprehensive overview of the existing literature on player styles and identify research areas required for further development, offering new directions for future research. Following the PRISMA guidelines for systematic reviews, we conducted a search using a specific strategy across four databases (PubMed, Scopus, Web of Science, and SPORTDiscus). Inclusion and exclusion criteria were applied to the initial search results, ultimately identifying twelve studies suitable for inclusion in this review. Through thematic analysis and qualitative evaluation of these studies, several key findings emerged: (a) a lack of a structured theoretical framework for player styles based on their positions within the team formation, (b) absence of studies investigating the influence of contextual variables on player styles, (c) methodological deficiencies observed in the reviewed studies, and (d) disparity in the objectives of sports science and data science studies. By identifying these gaps in the literature and presenting a structured framework for player styles (based on the compilation of all reported styles from the reviewed studies), this review aims to assist team stakeholders and provide guidance for future research endeavors.
RESUMO
As a result of social progress and improved living conditions, which have contributed to a prolonged life expectancy, the prevalence of strokes has increased and has become a significant phenomenon. Despite the available stroke treatment options, patients frequently suffer from significant disability after a stroke. Initial stroke severity is a significant predictor of functional dependence and mortality following an acute stroke. The current study aims to collect and analyze data from the hyperacute and acute phases of stroke, as well as from the medical history of the patients, in order to develop an explainable machine learning model for predicting stroke-related neurological deficits at discharge, as measured by the National Institutes of Health Stroke Scale (NIHSS). More specifically, we approached the data as a binary task problem: improvement of NIHSS progression vs. worsening of NIHSS progression at discharge, using baseline data within the first 72 h. For feature selection, a genetic algorithm was applied. Using various classifiers, we found that the best scores were achieved from the Random Forest (RF) classifier at the 15 most informative biomarkers and parameters for the binary task of the prediction of NIHSS score progression. RF achieved 91.13% accuracy, 91.13% recall, 90.89% precision, 91.00% f1-score, 8.87% FNrate and 4.59% FPrate. Those biomarkers are: age, gender, NIHSS upon admission, intubation, history of hypertension and smoking, the initial diagnosis of hypertension, diabetes, dyslipidemia and atrial fibrillation, high-density lipoprotein (HDL) levels, stroke localization, systolic blood pressure levels, as well as erythrocyte sedimentation rate (ESR) levels upon admission and the onset of respiratory infection. The SHapley Additive exPlanations (SHAP) model interpreted the impact of the selected features on the model output. Our findings suggest that the aforementioned variables may play a significant role in determining stroke patients' NIHSS progression from the time of admission until their discharge.
RESUMO
Knee Osteoarthritis (ΚΟΑ) is a degenerative joint disease of the knee that results from the progressive loss of cartilage. Due to KOA's multifactorial nature and the poor understanding of its pathophysiology, there is a need for reliable tools that will reduce diagnostic errors made by clinicians. The existence of public databases has facilitated the advent of advanced analytics in KOA research however the heterogeneity of the available data along with the observed high feature dimensionality make this diagnosis task difficult. The objective of the present study is to provide a robust Feature Selection (FS) methodology that could: (i) handle the multidimensional nature of the available datasets and (ii) alleviate the defectiveness of existing feature selection techniques towards the identification of important risk factors which contribute to KOA diagnosis. For this aim, we used multidimensional data obtained from the Osteoarthritis Initiative database for individuals without or with KOA. The proposed fuzzy ensemble feature selection methodology aggregates the results of several FS algorithms (filter, wrapper and embedded ones) based on fuzzy logic. The effectiveness of the proposed methodology was evaluated using an extensive experimental setup that involved multiple competing FS algorithms and several well-known ML models. A 73.55% classification accuracy was achieved by the best performing model (Random Forest classifier) on a group of twenty-one selected risk factors. Explainability analysis was finally performed to quantify the impact of the selected features on the model's output thus enhancing our understanding of the rationale behind the decision-making mechanism of the best model.
Assuntos
Osteoartrite do Joelho , Algoritmos , Lógica Fuzzy , Humanos , Articulação do Joelho , Aprendizado de Máquina , Osteoartrite do Joelho/diagnóstico por imagemRESUMO
BACKGROUND AND OBJECTIVE: Multi-atlas based segmentation techniques, which rely on an atlas library comprised of training images labeled by an expert, have proven their effectiveness in multiple automatic segmentation applications. However, the usage of exhaustive patch libraries combined with the voxel-wise labeling incur a large computational cost in terms of memory requirements and execution times. METHODS: To confront this shortcoming, we propose a novel two-stage multi-atlas approach designed under the Semi-Supervised Learning (SSL) framework. The main properties of our method are as follows: First, instead of the voxel-wise labeling approach, the labeling of target voxels is accomplished here by exploiting the spectral content of globally sampled datasets from the target image, along with their spatially correspondent data collected from the atlases. Following SSL, voxels classification is boosted by incorporating unlabeled data from the target image, in addition to the labeled ones from atlas library. Our scheme integrates constructively fruitful concepts, including sparse reconstructions of voxels from linear neighborhoods, HOG feature descriptors of patches/regions, and label propagation via sparse graph constructions. Segmentation of the target image is carried out in two stages: stage-1 focuses on the sampling and labeling of global data, while stage-2 undertakes the above tasks for the out-of-sample data. Finally, we propose different graph-based methods for the labeling of global data, while these methods are extended to deal with the out-of-sample voxels. RESULTS: A thorough experimental investigation is conducted on 76 subjects provided by the publicly accessible Osteoarthritis Initiative (OAI) repository. Comparative results and statistical analysis demonstrate that the suggested methodology exhibits superior segmentation performance compared to the existing patch-based methods, across all evaluation metrics (DSC:88.89%, Precision: 89.86%, Recall: 88.12%), while at the same time it requires a considerably reduced computational load (>70% reduction on average execution time with respect to other patch-based). In addition, our approach is favorably compared against other non patch-based and deep learning methods in terms of performance accuracy (on the 3-class problem). A final experimentation on a 5-class setting of the problems demonstrates that our approach is capable of achieving performance comparable to existing state-of-the-art knee cartilage segmentation methods (DSC:88.22% and DSC:85.84% for femoral and tibial cartilage respectively).
Assuntos
Cartilagem , Articulação do Joelho , Humanos , Articulação do Joelho/diagnóstico por imagem , Aprendizado de Máquina Supervisionado , Fêmur , TíbiaRESUMO
Stroke is an acute neurological dysfunction attributed to a focal injury of the central nervous system due to reduced blood flow to the brain. Nowadays, stroke is a global threat associated with premature death and huge economic consequences. Hence, there is an urgency to model the effect of several risk factors on stroke occurrence, and artificial intelligence (AI) seems to be the appropriate tool. In the present study, we aimed to (i) develop reliable machine learning (ML) prediction models for stroke disease; (ii) cope with a typical severe class imbalance problem, which is posed due to the stroke patients' class being significantly smaller than the healthy class; and (iii) interpret the model output for understanding the decision-making mechanism. The effectiveness of the proposed ML approach was investigated in a comparative analysis with six well-known classifiers with respect to metrics that are related to both generalization capability and prediction accuracy. The best overall false-negative rate was achieved by the Multi-Layer Perceptron (MLP) classifier (18.60%). Shapley Additive Explanations (SHAP) were employed to investigate the impact of the risk factors on the prediction output. The proposed AI method could lead to the creation of advanced and effective risk stratification strategies for each stroke patient, which would allow for timely diagnosis and the right treatments.
RESUMO
Coronavirus disease 2019 (COVID-19) has resulted in approximately 5 million deaths around the world with unprecedented consequences in people's daily routines and in the global economy. Despite vast increases in time and money spent on COVID-19-related research, there is still limited information about the factors at the country level that affected COVID-19 transmission and fatality in EU. The paper focuses on the identification of these risk factors using a machine learning (ML) predictive pipeline and an associated explainability analysis. To achieve this, a hybrid dataset was created employing publicly available sources comprising heterogeneous parameters from the majority of EU countries, e.g., mobility measures, policy responses, vaccinations, and demographics/generic country-level parameters. Data pre-processing and data exploration techniques were initially applied to normalize the available data and decrease the feature dimensionality of the data problem considered. Then, a linear ε-Support Vector Machine (ε-SVM) model was employed to implement the regression task of predicting the number of deaths for each one of the three first pandemic waves (with mean square error of 0.027 for wave 1 and less than 0.02 for waves 2 and 3). Post hoc explainability analysis was finally applied to uncover the rationale behind the decision-making mechanisms of the ML pipeline and thus enhance our understanding with respect to the contribution of the selected country-level parameters to the prediction of COVID-19 deaths in EU.
Assuntos
COVID-19 , COVID-19/epidemiologia , Europa (Continente)/epidemiologia , Humanos , Aprendizado de Máquina , Fatores de Risco , Máquina de Vetores de SuporteRESUMO
Anterior cruciate ligament (ACL) deficient and reconstructed knees display altered biomechanics during gait. Identifying significant gait changes is important for understanding normal and ACL function and is typically performed by statistical approaches. This paper focuses on the development of an explainable machine learning (ML) empowered methodology to: (i) identify important gait kinematic, kinetic parameters and quantify their contribution in the diagnosis of ACL injury and (ii) investigate the differences in sagittal plane kinematics and kinetics of the gait cycle between ACL deficient, ACL reconstructed and healthy individuals. For this aim, an extensive experimental setup was designed in which three-dimensional ground reaction forces and sagittal plane kinematic as well as kinetic parameters were collected from 151 subjects. The effectiveness of the proposed methodology was evaluated using a comparative analysis with eight well-known classifiers. Support Vector Machines were proved to be the best performing model (accuracy of 94.95%) on a group of 21 selected biomechanical parameters. Neural Networks accomplished the second best performance (92.89%). A state-of-the-art explainability analysis based on SHapley Additive exPlanations (SHAP) and conventional statistical analysis were then employed to quantify the contribution of the input biomechanical parameters in the diagnosis of ACL injury. Features, that would have been neglected by the traditional statistical analysis, were identified as contributing parameters having significant impact on the ML model's output for ACL injury during gait.
Assuntos
Lesões do Ligamento Cruzado Anterior , Ligamento Cruzado Anterior , Lesões do Ligamento Cruzado Anterior/diagnóstico , Fenômenos Biomecânicos , Marcha , Humanos , Articulação do Joelho , Aprendizado de MáquinaRESUMO
The improved treatment of knee injuries critically relies on having an accurate and cost-effective detection. In recent years, deep-learning-based approaches have monopolized knee injury detection in MRI studies. The aim of this paper is to present the findings of a systematic literature review of knee (anterior cruciate ligament, meniscus, and cartilage) injury detection papers using deep learning. The systematic review was carried out following the PRISMA guidelines on several databases, including PubMed, Cochrane Library, EMBASE, and Google Scholar. Appropriate metrics were chosen to interpret the results. The prediction accuracy of the deep-learning models for the identification of knee injuries ranged from 72.5-100%. Deep learning has the potential to act at par with human-level performance in decision-making tasks related to the MRI-based diagnosis of knee injuries. The limitations of the present deep-learning approaches include data imbalance, model generalizability across different centers, verification bias, lack of related classification studies with more than two classes, and ground-truth subjectivity. There are several possible avenues of further exploration of deep learning for improving MRI-based knee injury diagnosis. Explainability and lightweightness of the deployed deep-learning systems are expected to become crucial enablers for their widespread use in clinical practice.