Performance Of Model Research Articles

Predicting response to neoadjuvant chemotherapy for colorectal liver metastasis using deep learning on prechemotherapy cross-sectional imaging.

Deep learning models (DLMs) are applied across domains of health sciences to generate meaningful predictions. DLMs make use of neural networks to generate predictions from discrete data inputs. This study employs DLM on prechemotherapy cross-sectional imaging to predict patients' response to neoadjuvant chemotherapy. Adult patients with colorectal liver metastasis who underwent surgery after neoadjuvant chemotherapy were included. A DLM was trained on computed tomography images using attention-based multiple-instance learning. A logistic regression model incorporating clinical parameters of the Fong clinical risk score was used for comparison. Both model performances were benchmarked against the Response Evaluation Criteria in Solid Tumors criteria. A receiver operating curve was created and resulting area under the curve (AUC) was determined. Ninety-five patients were included, with 33,619 images available for study inclusion. Ninety-five percent of patients underwent 5-fluorouracil-based chemotherapy with oxaliplatin and/or irinotecan. Sixty percent of the patients were categorized as chemotherapy responders (30% reduction in tumor diameter). The DLM had an AUC of 0.77. The AUC for the clinical model was 0.41. Image-based DLM for prediction of response to neoadjuvant chemotherapy in patients with colorectal cancer liver metastases was superior to a clinical-based model. These results demonstrate potential to identify nonresponders to chemotherapy and guide select patients toward earlier curative resection.

A novel artificial intelligence-based endoscopic ultrasonography diagnostic system for diagnosing the invasion depth of early gastric cancer.

We developed an artificial intelligence (AI)-based endoscopic ultrasonography (EUS) system for diagnosing the invasion depth of early gastric cancer (EGC), and we evaluated the performance of this system. A total of 8280 EUS images from 559 EGC cases were collected from 11 institutions. Within this dataset, 3451 images (285 cases) from one institution were used as a development dataset. The AI model consisted of segmentation and classification steps, followed by the CycleGAN method to bridge differences in EUS images captured by different equipment. AI model performance was evaluated using an internal validation dataset collected from the same institution as the development dataset (1726 images, 135 cases). External validation was conducted using images collected from the other 10 institutions (3103 images, 139 cases). The area under the curve (AUC) of the AI model in the internal validation dataset was 0.870 (95% CI: 0.796-0.944). Regarding diagnostic performance, the accuracy/sensitivity/specificity values of the AI model, experts (n = 6), and nonexperts (n = 8) were 82.2/63.4/90.4%, 81.9/66.3/88.7%, and 68.3/60.9/71.5%, respectively. The AUC of the AI model in the external validation dataset was 0.815 (95% CI: 0.743-0.886). The accuracy/sensitivity/specificity values of the AI model (74.1/73.1/75.0%) and the real-time diagnoses of experts (75.5/79.1/72.2%) in the external validation dataset were comparable. Our AI model demonstrated a diagnostic performance equivalent to that of experts.

T2-weighted imaging-based deep-learning method for noninvasive prostate cancer detection and Gleason grade prediction: a multicenter study.

To noninvasively detect prostate cancer and predict the Gleason grade using single-modality T2-weighted imaging with a deep-learning approach. Patients with prostate cancer, confirmed by histopathology, who underwent magnetic resonance imaging examinations at our hospital during September 2015-June 2022 were retrospectively included in an internal dataset. An external dataset from another medical center and a public challenge dataset were used for external validation. A deep-learning approach was designed for prostate cancer detection and Gleason grade prediction. The area under the curve (AUC) was calculated to compare the model performance. For prostate cancer detection, the internal datasets comprised data from 195 healthy individuals (age: 57.27 ± 14.45 years) and 302 patients (age: 72.20 ± 8.34 years) diagnosed with prostate cancer. The AUC of our model for prostate cancer detection in the validation set (n = 96, 19.7%) was 0.918. For Gleason grade prediction, datasets comprising data from 283 of 302 patients with prostate cancer were used, with 227 (age: 72.06 ± 7.98 years) and 56 (age: 72.78 ± 9.49 years) patients being used for training and testing, respectively. The external and public challenge datasets comprised data from 48 (age: 72.19 ± 7.81 years) and 91 patients (unavailable information on age), respectively. The AUC of our model for Gleason grade prediction in the training set (n = 227) was 0.902, whereas those of the validation (n = 56), external validation (n = 48), and public challenge validation sets (n = 91) were 0.854, 0.776, and 0.838, respectively. Through multicenter dataset validation, our proposed deep-learning method could detect prostate cancer and predict the Gleason grade better than human experts. Precise prostate cancer detection and Gleason grade prediction have great significance for clinical treatment and decision making. Prostate segmentation is easier to annotate than prostate cancer lesions for radiologists. Our deep-learning method detected prostate cancer and predicted the Gleason grade, outperforming human experts. Non-invasive Gleason grade prediction can reduce the number of unnecessary biopsies.

Error detection for radiotherapy planning validation based on deep learning networks.

Quality assurance (QA) of patient-specific treatment plans for intensity-modulated radiation therapy (IMRT) and volumetric modulated arc therapy (VMAT) necessitates prior validation. However, the standard methodology exhibits deficiencies and lacks sensitivity in the analysis of positional dose distribution data, leading to difficulties in accurately identifying reasons for plan verification failure. This issue complicates and impedes the efficiency of QA tasks. The primary aim of this research is to utilize deep learning algorithms for the extraction of 3D dose distribution maps and the creation of a predictive model for error classification across multiple machine models, treatment methodologies, and tumor locations. We devised five categories of validation plans (normal, gantry error, collimator error, couch error, and dose error), conforming to tolerance limits of different accuracy levels and employing 3D dose distribution data from a sample of 94 tumor patients. A CNN model was then constructed to predict the diverse error types, with predictions compared against the gamma pass rate (GPR) standard employing distinct thresholds (3%, 3mm; 3%, 2mm; 2%, 2mm) to evaluate the model's performance. Furthermore, we appraised the model's robustness by assessing its functionality across diverse accelerators. The accuracy, precision, recall, and F1 scores of CNN model performance were 0.907, 0.925, 0.907, and 0.908, respectively. Meanwhile, the performance on another device is 0.900, 0.918, 0.900, and 0.898. In addition, compared to the GPR method, the CNN model achieved better results in predicting different types of errors. When juxtaposed with the GPR methodology, the CNN model exhibits superior predictive capability for classification in the validation of the radiation therapy plan on different devices. By using this model, the plan validation failures can be detected more rapidly and efficiently, minimizing the time required for QA tasks and serving as a valuable adjunct to overcome the constraints of the GPR method.

Multimodal radiotherapy dose prediction using a multi-task deep learning model.

In radiation therapy (RT), accelerated partial breast irradiation (APBI) has emerged as an increasingly preferred treatment modality over conventional whole breast irradiation due to its targeted dose delivery and shorter course of treatment. APBI can be delivered through various modalities including Cobalt-60-based systems and linear accelerators with C-arm, O-ring, or robotic arm design. Each modality possesses distinct features, such as beam energy or the degrees of freedom in treatment planning, which influence their respective dose distributions. These modality-specific considerations emphasize the need for a quantitative approach in determining the optimal dose delivery modality on a patient-specific basis. However, manually generating treatment plans for each modality across every patient is time-consuming and clinically impractical. We aim to develop an efficient and personalized approach for determining the optimal RT modality for APBI by training predictive models using two different deep learning-based convolutional neural networks. The baseline network performs a single-task (ST), predicting dose for a single modality. Our proposed multi-task (MT) network, which is capable of leveraging shared information among different tasks, can concurrently predict dose distributions for various RT modalities. Utilizing patient-specific input data, such as a patient's computed tomography (CT) scan and treatment protocol dosimetric goals, the MT model predicts patient-specific dose distributions across all trained modalities. These dose distributions provide patients and clinicians quantitative insights, facilitating informed and personalized modality comparison prior to treatment planning. The dataset, comprising 28 APBI patients and their 92 treatment plans, was partitioned into training, validation, and test subsets. Eight patients were dedicated to the test subset, leaving 68 treatment plans across 20 patients to divide between the training and validation subsets. ST models were trained for each modality, and one MT model was trained to predict doses for all modalities simultaneously. Model performance was evaluated across the test dataset in terms of Mean Absolute Percent Error (MAPE). We conducted statistical analysis of model performance using the two-tailed Wilcoxon signed-rank test. Training times for five ST models ranged from 255 to 430min per modality, totaling 1925min, while the MT model required 2384min. MT model prediction required an average of 1.82s per patient, compared to ST model predictions at 0.93s per modality. The MT model yielded MAPE of 1.1033±0.3627% as opposed to the collective MAPE of 1.2386±0.3872% from ST models, and the differences were statistically significant (p=0.0003, 95% confidence interval=[-0.0865, -0.0712]). Our study highlights the potential benefits of a MT learning framework in predicting RT dose distributions across various modalities without notable compromises. This MT architecture approach offers several advantages, such as flexibility, scalability, and streamlined model management, making it an appealing solution for clinical deployment. With such a MT model, patients can make more informed treatment decisions, physicians gain more quantitative insight for pre-treatment decision-making, and clinics can better optimize resource allocation. With our proposed goal array and MT framework, we aim to expand this work to a site-agnostic dose prediction model, enhancing its generalizability and applicability.

Biomarkers to improve functional outcome prediction after ischemic stroke: Results from the SICFAIL, STRAWINSKI, and PREDICT studies.

Acute ischemic stroke (AIS) outcome prognostication remains challenging despite available prognostic models. We investigated whether a biomarker panel improves the predictive performance of established prognostic scores. We investigated the improvement in discrimination, calibration, and overall performance by adding five biomarkers (procalcitonin, copeptin, cortisol, mid-regional pro-atrial natriuretic peptide (MR-proANP), and N-terminal pro-B-type natriuretic peptide (NT-proBNP)) to the Acute Stroke Registry and Analysis of Lausanne (ASTRAL) and age/NIHSS scores using data from two prospective cohort studies (SICFAIL, PREDICT) and one clinical trial (STRAWINSKI). Poor outcome was defined as mRS > 2 at 12 (SICFAIL, derivation dataset) or 3 months (PREDICT/STRAWINSKI, pooled external validation dataset). Among 412 SICFAIL participants (median age 70 years, quartiles 59-78; 63% male; median NIHSS score 3, quartiles 1-5), 29% had a poor outcome. Area under the curve of the ASTRAL and age/NIHSS were 0.76 (95% CI 0.71-0.81) and 0.77 (95% CI 0.73-0.82), respectively. Copeptin (0.79, 95% CI 0.74-0.84), NT-proBNP (0.80, 95% CI 0.76-0.84), and MR-proANP (0.79, 95% CI 0.75-0.84) significantly improved ASTRAL score's discrimination, calibration, and overall performance. Copeptin improved age/NIHSS model's discrimination, copeptin, MR-proANP, and NT-proBNP improved its calibration and overall performance. In the validation dataset (450 patients, median age 73 years, quartiles 66-81; 54% men; median NIHSS score 8, quartiles 3-14), copeptin was independently associated with various definitions of poor outcome and also mortality. Copeptin did not increase model's discrimination but it did improve calibration and overall model performance. Copeptin, NT-proBNP, and MR-proANP improved modest but consistently the predictive performance of established prognostic scores in patients with mild AIS. Copeptin was most consistently associated with poor outcome in patients with moderate to severe AIS, although its added prognostic value was less obvious.

A novel nomogram for the prediction of perforation during endoscopic submucosal dissection for colorectal neoplasms.

High perforation risk hinders the widespread adoption of ESD for colorectal neoplasms. This study was performed to determine the risk factors of colorectal endoscopic submucosal dissection (ESD)-induced perforation and develop a predictive model. A total of 1046 colorectal neoplasms in 1011 patients were retrospectively enrolled from January 2011 to December 2021, in a single tertiary center as the derivation cohort. We identified independent risk factors for perforation using univariate analysis and multi-variate logistic regression. A nomogram was developed based on the logistic regression model and prospectively applied to 266 colorectal neoplasms as the validation cohort. The performance of the predictive model was evaluated with the receiver operating characteristic curve, calibration plot, and decision curve analysis. Independent pre-operative factors for colorectal ESD-induced perforation were tumor located in the left colon [odds ratio (OR) 2.39, P = 0.040], size ≥ 40 mm (OR 3.36, P < 0.001), ≥2/3 circumference (OR 7.55, P = 0.004), located across folds (OR 6.26, P < 0.001), and laterally spreading tumor (non-granular type, OR 2.34, P = 0.029; granular type, OR 2.46, P = 0.021). The nomogram model incorporating the pre-operative factors performed well in both the derivation and validation cohorts (areas under the curve of 0.750 and 0.806, respectively). Decision curve analysis demonstrated that the clinical benefit of the nomogram was favorable. The novel nomogram, developed and prospectively validated, incorporating tumor size, location, and morphology can successfully predict perforation during ESD for colorectal neoplasms.

RT: a Retrieving and Chain-of-Thought framework for few-shot medical named entity recognition.

This article aims to enhance the performance of larger language models (LLMs) on the few-shot biomedical named entity recognition (NER) task by developing a simple and effective method called Retrieving and Chain-of-Thought (RT) framework and to evaluate the improvement after applying RT framework. Given the remarkable advancements in retrieval-based language model and Chain-of-Thought across various natural language processing tasks, we propose a pioneering RT framework designed to amalgamate both approaches. The RT approach encompasses dedicated modules for information retrieval and Chain-of-Thought processes. In the retrieval module, RT discerns pertinent examples from demonstrations during instructional tuning for each input sentence. Subsequently, the Chain-of-Thought module employs a systematic reasoning process to identify entities. We conducted a comprehensive comparative analysis of our RT framework against 16 other models for few-shot NER tasks on BC5CDR and NCBI corpora. Additionally, we explored the impacts of negative samples, output formats, and missing data on performance. Our proposed RT framework outperforms other LMs for few-shot NER tasks with micro-F1 scores of 93.50 and 91.76 on BC5CDR and NCBI corpora, respectively. We found that using both positive and negative samples, Chain-of-Thought (vs Tree-of-Thought) performed better. Additionally, utilization of a partially annotated dataset has a marginal effect of the model performance. This is the first investigation to combine a retrieval-based LLM and Chain-of-Thought methodology to enhance the performance in biomedical few-shot NER. The retrieval-based LLM aids in retrieving the most relevant examples of the input sentence, offering crucial knowledge to predict the entity in the sentence. We also conducted a meticulous examination of our methodology, incorporating an ablation study. The RT framework with LLM has demonstrated state-of-the-art performance on few-shot NER tasks.

Multiphysics performance of surrogate models on skewed pole surface-mounted permanent magnet motors

Purpose Evaluating the multiphysics performance of an electric motor can be a computationally intensive process, especially where several complex subsystems of the motor are coupled together. For example, evaluating acoustic noise requires the coupling of the electromagnetic, structural and acoustic models of the electric motor. Where skewed poles are considered in the design, the problem becomes a purely three-dimensional (3D) multiphysics problem, which could increase the computational burden astronomically. This study, therefore, aims to introduce surrogate models in the design process to reduce the computational cost associated with solving such 3D-coupled multiphysics problems. Design/methodology/approach The procedure involves using the finite element (FE) method to generate a database of several skewed rotor pole surface-mounted permanent magnet synchronous motors and their corresponding electromagnetic, structural and acoustic performances. Then, a surrogate model is fitted to the data to generate mapping functions that could be used in place of the time-consuming FE simulations. Findings It was established that the surrogate models showed promising results in predicting the multiphysics performance of skewed pole surface-mounted permanent magnet motors. As such, such models could be used to handle the skewing aspects, which has always been a major design challenge due to the scarcity of simulation tools with stepwise skewing capability. Originality/value The main contribution involves the use of surrogate models to replace FE simulations during the design cycle of skewed pole surface-mounted permanent magnet motors without compromising the integrity of the electromagnetic, structural, and acoustic results of the motor.

Can geomorphic flood descriptors coupled with machine learning models enhance in quantifying flood risks over data-scarce catchments? Development of a hybrid framework for Ganga basin (India).

Quantifying flood risks through a cascade of hydraulic-cum-hydrodynamic modelling is data-intensive and computationally demanding- a major constraint for economically struggling and data-scarce low and middle-income nations. Under such circumstances, geomorphic flood descriptors (GFDs), that encompass the hidden characteristics of flood propensity may assist in developing a nuanced understanding of flood risk management. In line with this, the present study proposes a novel framework for estimating flood hazard and population exposure by leveraging GFDs and Machine Learning (ML) models over severely flood-prone Ganga basin. The study incorporates SHapley Additive exPlanations (SHAP) values in flood hazard modeling to justify the degree of influence of each GFD on the simulated floodplain maps. A set of 15 relevant GFDs derived from high-resolution CartoDEM are forced to five state-of-the-art ML models; AdaBoost, Random Forest, GBDT, XGBoost, and CatBoost, for predicting flood extents and depths. To enumerate the performance of ML models, a set of twelve statistical metrics are considered. Our result indicates a superior performance of XGBoost (κ = 0.72 and KGE = 82%) over other ML models in flood extent and flood depth prediction, resulting in about 47% of the population exposure to high-flood risks. The SHAP summary plots reveal a pre-dominance of Height Above Nearest Drainage during flood depth prediction. The study contributes significantly in comprehending our understanding of catchment characteristics and its influence in the process of sustainable disaster risk reduction. The results obtained from the study provide valuable recommendations for efficient flood management and mitigation strategies, especially over global data-scarce flood-prone basins.

Intraoperative quantitative crystalline lens nuclear opacities analysis based on crystalline lenSx platform

PurposeThe main objective is to quantify the lens nuclear opacity using spectral-domain optical coherence tomography (SD-OCT) and to evaluate its association with Lens Opacities Classification System III (LOCS-III) system, lens thickness (LT), and surgical parameters. The secondary objective is to assess the diagnostic model performance for hard nuclear cataract.MethodsThis study included 70 eyes of 57 adults with cataract, with 49 (70%) and 21 (30%) in training and validation cohort, respectively. Correlations of the average nuclear density (AND) /maximum nuclear density (MND) with LOCS-III scores, LT, and surgical parameters were analyzed. Univariate and multivariate logistic regression analysis, receiver operating characteristic curves and calibration curves were performed for the diagnostic of hard nuclear cataract.ResultsThe pre-operative uncorrected distance visual acuity (UDVA), intraocular pressure (IOP), mean axial length (AL), and LT were 1.20 ± 0.47 log MAR, 15.50 ± 2.87 mmHg, 27.34 ± 3.77 mm and 4.32 ± 0.45 mm, respectively. The average nuclear opalescence (NO) and nuclear colour (NC) scores were 3.61 ± 0.94 and 3.50 ± 0.91 (ranging from 1.00 to 6.90), respectively. The average AND and MND were 137.94 ± 17.01 and 230.01 ± 8.91, respectively. NC and NO scores both significantly correlated with the AND (rNC = 0.733, p = 0.000; rNO = 0.755, p = 0.000) and MND (rNC = 0.643, p = 0.000; rNO = 0.634, p = 0.000). In the training cohort, the area under the curve (AUC) of the model was 0.769 (P < 0.001, 95%CI 0.620–0.919), which had a good degree of differentiation (Fig. 2a). The calibration curve showed good agreement between predicted and actual probability.ConclusionThe nuclear density measurement on SD-OCT images can serve as an objective and reliable indicator for quantifying nuclear density.

Multi-year soundscape recordings and automated call detection reveals varied impact of moonlight on calling activity of neotropical forest katydids.

Night-time light can have profound ecological effects, even when the source is natural moonlight. The impacts of light can, however, vary substantially by taxon, habitat and geographical region. We used a custom machine learning model built with the Python package Koogu to investigate the in situ effects of moonlight on the calling activity of neotropical forest katydids over multiple years. We prioritised species with calls that were commonly detected in human annotated data, enabling us to evaluate model performance. We focused on eight species of katydids that the model identified with high precision (generally greater than 0.90) and moderate-to-high recall (minimum 0.35), ensuring that detections were generally correct and that many calls were detected. These results suggest that moonlight has modest effects on the amount of calling, with the magnitude and direction of effect varying by species: half of the species showed positive effects of light and half showed negative. These findings emphasize the importance of understanding natural history for anticipating how biological communities respond to moonlight. The methods applied in this project highlight the emerging opportunities for evaluating large quantities of data with machine learning models to address ecological questions over space and time. This article is part of the theme issue 'Towards a toolkit for global insect biodiversity monitoring'.

Analysing LULC transformations using remote sensing data: insights from a multilayer perceptron neural network approach

ABSTRACT The study examines the complex dynamics of changes in LULC over three decades, focused on the years 1992, 2002, 2012, and 2022. The research highlights the significance of comprehending these alterations within the framework of environmental and socio-economic consequences. The changes in land use and land cover (LULC) have significant and far-reaching effects on ecosystems, biodiversity, and human livelihoods. This study offers useful information for politicians, conservationists, and urban planners by examining historical patterns and forecasting future changes. The study utilized a Multilayer Perceptron Neural Network (MLP-NN), a well-known machine learning technique that excels at collecting intricate patterns. This model’s design had three layers: input, hidden, and output. The model underwent 10,000 iterations during its training process, and a thorough statistical analysis was conducted to assess the impact of each driving component. The MLP-NN model demonstrated impressive performance, with a skill measure of 0.8724 and an accuracy rate of 89.08%. The accuracy of the LULC estimates for 2022 was verified by comparing them with observed data, ensuring the model’s reliability. Moreover, the presence of evidence likely was found to be a significant factor that had a substantial impact on the accuracy of the model. The study highlights the effectiveness of the MLP-NN model in accurately predicting changes in LULC. The model’s exceptional accuracy and proficiency make it a powerful tool for future LULC forecasts. Identifying the primary causes of model performance and understanding their implications may help to enhance land management strategies, encourage spatial planning, guide accurate decision-making, and facilitate the development of policies that align with sustainable growth and development.

Interpretable machine learning based on CT-derived extracellular volume fraction to predict pathological grading of hepatocellular carcinoma.

To develop a non-invasive auxiliary assessment method based on CT-derived extracellular volume (ECV) to predict the pathological grading (PG) of hepatocellular carcinoma (HCC). The study retrospectively analyzed 238 patients who underwent HCC resection surgery between January 2013 and April 2023. Six machine learning algorithms were employed to construct predictive models for HCC PG: logistic regression, extreme gradient boosting, Light Gradient Boosting Machine (LightGBM), random forest, adaptive boosting, and Gaussian naive Bayes. Model performance was evaluated using receiver operating characteristic curve analysis, including area under the curve (AUC), sensitivity, specificity, accuracy, positive predictive value, negative predictive value, and F1 score. Calibration plots were used for visual evaluation of model calibration. Clinical decision curve analysis was performed to assess potential clinical utility by calculating net benefit. 166 patients from Hospital A were allocated to the training set, while 72 patients from Hospital B (constituting 30.25% of the total sample) were assigned to the test set. The model achieved an AUC of 1.000 (95%CI: 1.000-1.000) in the training set and 0.927 (95%CI: 0.837-0.999) in the validation set, respectively. Ultimately, the model achieved an AUC of 0.909 (95%CI: 0.837-0.980) in the test set, with an accuracy of 0.778, sensitivity of 0.906, specificity of 0.789, negative predictive value of 0.556, and F1 score of 0.908. This study successfully developed and validated a non-invasive auxiliary assessment method based on CT-derived ECV to predict the HCC PG, providing important supplementary information for clinical decision-making.

Assessment of Roughness Characterization Methods for Data-Driven Predictions

AbstractA comparative analysis is undertaken to explore the impact of various roughness characterization methods as input variables on the performance of data-driven predictive models for estimating the roughness equivalent sand-grain size $$k_s$$ k s . The first type of model, denoted as $$\text {ENN}_\text {PS}$$ ENN PS , incorporates the roughness height probability density function (p.d.f.) and power spectrum (PS), while the second type of model, $$\text {ENN}_\text {PA}$$ ENN PA , utilizes a finite set of 17 roughness statistical parameters as input variables. Furthermore, a simplified parameter-based model, denoted as $$\text {ENN}_\text {PAM}$$ ENN PAM , is considered, which features only 6 input roughness parameters. The models are trained based on identical databases and evaluated using roughness samples similar to the training databases as well as an external testing database based on literature. While the predictions based on p.d.f. and PS achieves a stable error level of around 10% among all considered testing samples, a notable deterioration in performance is observed for the parameter-based models for the external testing database, indicating a lower extrapolating capability to diverse roughness types. Finally, the sensitivity analysis on different types of roughness confirms an effective identification of distinct roughness effects by $$\text {ENN}_\text {PAM}$$ ENN PAM , which is not observed for $$\text {ENN}_\text {PA}$$ ENN PA . We hypothesize that the successful training of $$\text {ENN}_\text {PAM}$$ ENN PAM is attributed to the enhanced training efficiency linked to the lower input dimensionality.

A comparative study of pre-processing methods to improve glioma segmentation performance in brain MRI using deep learning

Background: Glioma is the most common brain tumor in adult patients and requires accurate treatment. The delineation of tumor boundaries must be accurate and precise, which is crucial for treatment planning. Currently, delineating boundaries for tumors is a tedious, time-consuming task and may be prone to human error among oncologists. Therefore, artificial intelligence plays a vital role in reducing these problems. Objective: This study aims to find a relationship between improving image enhancement and evaluating the performance of deep learning models for segmenting glioma image data on brain MRI images. Materials and methods: The BraTs2023 dataset was used in this study. The image dataset was converted from three dimensions to two dimensions and then subjected to pre-processing via four image enhancement techniques, including contrast-limited adaptive histogram equalization (CLAHE), gamma correction (GC), non-local mean filter (NLMF), and median and Wiener filter (MWF). Subsequently, it was evaluated for structural similarity index (SSIM) and mean squared error. The deep learning segmentation model was created using the U-Net architecture and assessed for dice similarity coefficient (DSC), accuracy, precision, recall, F1-score, and Jaccard index for tumor segmentation. Results: The performance of enhanced image results for CLAHE, GC, NLMF, and MWF techniques shows SSIM values of 0.912, 0.905, 0.999, and 0.911, respectively. The dice similarity coefficient (DSC) for segmentation without image enhancement was 0.817. The DSC of segmentation for CLAHE, GC, NLMF, and MWF techniques were 0.818, 0.812, 0.820, and 0.797, respectively. Conclusion: The enhanced image technique could affect the performance of tumor segmentation. by the enhanced image for use in a trained model may increase or decrease performance depending on the chosen image enhancement technique and the parameters determined by each method.

Xai-driven knowledge distillation of large language models for efficient deployment on low-resource devices

Large Language Models (LLMs) are characterized by their inherent memory inefficiency and compute-intensive nature, making them impractical to run on low-resource devices and hindering their applicability in edge AI contexts. To address this issue, Knowledge Distillation approaches have been adopted to transfer knowledge from a complex model, referred to as the teacher, to a more compact, computationally efficient one, known as the student. The aim is to retain the performance of the original model while substantially reducing computational requirements. However, traditional knowledge distillation methods may struggle to effectively transfer crucial explainable knowledge from an LLM teacher to the student, potentially leading to explanation inconsistencies and decreased performance. This paper presents DiXtill, a method based on a novel approach to distilling knowledge from LLMs into lightweight neural architectures. The main idea is to leverage local explanations provided by an eXplainable Artificial Intelligence (XAI) method to guide the cross-architecture distillation of a teacher LLM into a self-explainable student, specifically a bi-directional LSTM network.Experimental results show that our XAI-driven distillation method allows the teacher explanations to be effectively transferred to the student, resulting in better agreement compared to classical distillation methods,thus enhancing the student interpretability. Furthermore, it enables the student to achieve comparable performance to the teacher LLM while also delivering a significantly higher compression ratio and speedup compared to other techniques such as post-training quantization and pruning, which paves the way for more efficient and sustainable edge AI applications

Performance Evaluation of VGG_19 Model for Brain Tumor Datasets

This paper presents a comprehensive assessment of the VGG19 model for detecting brain tumors using deep learning methods, particularly convolutional neural networks (CNNs). The study evaluates key performance metrics such as accuracy, precision, recall, F1 score, and area under the curve (AUC). The investigation delves into the influence of different activation functions, learning rates, and epochs on the VGG19 model's performance. Activation functions are crucial for feature extraction and model convergence, with a comparison made between ReLU and Leaky ReLU to determine their effectiveness in enhancing brain tumor detection accuracy. Additionally, the study examines the impact of various learning rates (0.01, 0.001, 0.0001) on the model's convergence speed and overall performance. Different numbers of training epochs (15, 25, 50) are also considered to strike a balance between computational efficiency and model effectiveness. Evaluation metrics encompass accuracy, precision, recall, F1 score, and AUC, providing a comprehensive assessment of the model's performance across different configurations. The goal is to identify optimal combinations of activation functions, learning rates, and epochs to maximize the VGG19 model's accuracy in detecting brain tumors. This research contributes to advancing efficient and accurate methodologies for brain tumor detection, facilitating early diagnosis and treatment planning. The insights gained from this study can guide the development of more effective deep learning-based systems for medical image analysis, benefiting both patients and healthcare professionals.

Predicting Bus Travel Time in Cheonan City Through Deep Learning Utilizing Digital Tachograph Data

Urban transportation systems are increasingly burdened by traffic congestion, a consequence of population growth and heightened reliance on private vehicles. This congestion not only disrupts travel efficiency but also undermines productivity and urban resident’s overall well-being. A critical step in addressing this challenge is the accurate prediction of bus travel times, which is essential for mitigating congestion and improving the experience of public transport users. To tackle this issue, this study introduces the Hybrid Temporal Forecasting Network (HTF-NET) model, a framework that integrates machine learning techniques. The model combines an attention mechanism with Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) layers, enhancing its predictive capabilities. Further refinement is achieved through a Support Vector Regressor (SVR), enabling the generation of precise bus travel time predictions. To evaluate the performance of the HTF-NET model, comparative analyses are conducted with six deep learning models using real-world digital tachograph (DTG) data obtained from intracity buses in Cheonan City, South Korea. These models includes various architectures, including different configurations of LSTM and GRU, such as bidirectional and stacked architectures. The primary focus of the study is on predicting travel times from the Namchang Village bus stop to the Dongnam-gu Public Health Center, a crucial route in the urban transport network. Various experimental scenarios are explored, incorporating overall test data, and weekday and weekend data, with and without weather information, and considering different route lengths. Comparative evaluations against a baseline ARIMA model underscore the performance of the HTF-NET model. Particularly noteworthy is the significant improvement in prediction accuracy achieved through the incorporation of weather data. Evaluation metrics, including root mean squared error (RMSE), mean absolute error (MAE), and mean squared error (MSE), consistently highlight the superiority of the HTF-NET model, outperforming the baseline ARIMA model by a margin of 63.27% in terms of the RMSE. These findings provide valuable insights for transit agencies and policymakers, facilitating informed decisions regarding the management and optimization of public transportation systems.

Sign Speak: Recogninzing Sign Language with Machine Learning

Sign language serves as a critical means of communication for individuals with hearing impairments, enabling them to integrate into society effectively and express themselves. However, interpreting and recognizing sign language gestures present unique challenges due to the dynamic nature of gestures and spatial dependencies inherent in sign language communication. As a response, the SignSpeak project employs advanced machine learning techniques to address these challenges and enhance accessibility for the deaf and hard of hearing community. The project leverages a diverse dataset sourced from Kaggle, comprising images of sign language gestures captured in various contexts. The integration of advanced algorithms, such as 3D Convolutional Neural Networks (CNNs), and Gated Recurrent Units (GRUs), enables SignSpeak to recognize and interpret sign language gestures accurately and in real-time. This integration allows the model to capture both spatial and temporal features inherent in sign language, thus enabling more robust and accurate recognition. The project encompasses several critical stages, including data preprocessing, model development, training, and evaluation. Data preprocessing involves converting the image data into a suitable format and applying augmentation techniques to enhance the diversity and robustness of the dataset. Model development entails designing a deep learning architecture that combines CNNs and GRUs to effectively capture spatial and temporal dependencies in sign language gestures. Training the model involves optimizing parameters and hyperparameters to achieve optimal performance. Evaluation metrics such as accuracy, F1 score, and recall are utilized to assess the model's performance on both training and validation datasets. The trained model is then tested on a separate test dataset to evaluate its real- world performance and generalization ability. Experimental results demonstrate the efficacy of the SignSpeak approach in accurately recognizing and interpreting sign language gestures. The model achieves high accuracy scores, demonstrating its potential to enhance accessibility and inclusion for individuals with hearing impairments. By providing real-time translation of sign language into text or speech, SignSpeak contributes to breaking down communication barriers and promoting equal participation for all members of society.