Train Machine Learning Algorithms Research Articles

Machine Learning to Predict Outcomes of Fetal Cardiac Disease: A Pilot Study.

Prediction of outcomes following a prenatal diagnosis of congenital heart disease (CHD) is challenging. Machine learning (ML) algorithms may be used to reduce clinical uncertainty and improve prognostic accuracy. We performed a pilot study to train ML algorithms to predict postnatal outcomes based on clinical data. Specific objectives were to predict (1) in utero or neonatal death, (2) high-acuity neonatal care and (3) favorable outcomes. We included all fetuses with cardiac disease at Sunnybrook Health Sciences Centre, Toronto, Canada, from 2012 to 2021. Prediction models were created using the XgBoost algorithm (tree-based) with fivefold cross-validation. Among 211 cases of fetal cardiac disease, 61 were excluded (39 terminations, 21 lost to follow-up, 1 isolated arrhythmia), leaving a cohort of 150 fetuses. Fifteen (10%) demised (10 neonates) and 65 (48%) of live births required high acuity neonatal care. Of those with clinical follow-up, 60/87 (69%) had a favorable outcome. Prediction models for fetal or neonatal death, high acuity neonatal care and favorable outcome had AUCs of 0.76, 0.84 and 0.73, respectively. The most important predictors for death were the presence of non-cardiac abnormalities combined with more severe CHD. High acuity of postnatal care was predicted by anti Ro antibody and more severe CHD. Favorable outcome was most predicted by no right heart disease combined with genetic abnormalities, and maternal medications. Prediction models using ML provide good discrimination of key prenatal and postnatal outcomes among fetuses with congenital heart disease.

Simulation of gamma spectra from soil samples by using MCNP: A case study

A comparison between the MCNP simulated gamma spectrums based on the nuclear data from NuDat with the current version 3.0 and Nucléide-Lara against measured spectra based on the IAEA reference samples has been performed to assess the influence of nuclear data of photon decay processes and cross-section of photon interactions on the quality of the simulated spectra. As the results, the appearance of some abnormal energy peaks; namely 215 keV, 571 keV, 675 keV, 1227 keV of the NuDat-based simulated spectra and 90 keV, 94 keV, 106 keV, 416 keV of the Nucléide-Lara-based simulated spectra, which were present in neither the measured spectrum nor remaining simulated spectra, indicating issues with accuracy and completeness of these dataset. In addition, the good correlation between the combined dataset-based simulated spectra and reference samples-based measured spectra within the range of 50 keV to 2620 keV suggests that this MCNP simulation configuration can be used to generate a large simulated dataset for Machine Learning (ML) models that automatically identify and qualify radioactive isotope from gamma-ray spectra, overcoming the practical limitation of number of reference samples to sufficiently generate data for training and testing ML algorithms in the field of environmental radiation [1].

Bayesian data-driven framework for structural health monitoring of composite structures under limited experimental data

Ultrasonic-guided waves can be used to monitor the health of thin-walled structures. However, the run of experimental damage tests on materials like carbon fiber-reinforced plastics can be impractical and costly. Instead, numerical models can be used to create hybrid datasets to train machine learning algorithms, integrating data from numerical and experimental tests. This paper presents a Bayesian-driven framework to compensate for limited experimental data regarding Lamb wave propagation in composite plates. Using Bayesian inference, the framework updates a numerical finite element model, considering observed uncertainties by sampling posterior probability density functions for input parameters using Markov–Chain Monte Carlo simulations with the Metropolis-Hastings algorithm. A neural network surrogate model speeds-up these simulations, leading to a model that replicates the uncertain experimental setup. This model then generates data to augment true experimental data. Finally, a one-dimensional convolutional neural network is trained on a three different datasets to analyze Lamb wave signals and assess damage. Comparing training strategies shows the hybrid dataset augmented by samples generated by the updated FE model gives the most accurate damage size predictions.

High-speed camera system for efficient monitoring of invasive plant species along roadways

Invasive plant species pose ecological threats to native ecosystems, particularly in areas adjacent to roadways, considering that roadways represent lengthy corridors through which invasive species can propagate. Traditional manual survey methods for monitoring invasive plants are labor-intensive and limited in coverage. This paper introduces a high-speed camera system, named CamAlien, designed to be mounted on vehicles for efficient invasive plant species monitoring along roadways. The camera system captures high-quality images at rapid intervals, to monitor the full roadside when following traffic speed. The system utilizes a global shutter sensor to reduce distortion and geotagging for precise localistion. The camera system makes it possible to collect extensive data sets, which can be used for a digital library of the invasive species and their locations, but also subsequent training of machine learning algorithms for automated species recognition.

Toward Self-Driven Autonomous Material and Device Acceleration Platforms (AMADAP) for Emerging Photovoltaics Technologies.

ConspectusIn the ever-increasing renewable-energy demand scenario, developing new photovoltaic technologies is important, even in the presence of established terawatt-scale silicon technology. Emerging photovoltaic technologies play a crucial role in diversifying material flows while expanding the photovoltaic product portfolio, thus enhancing security and competitiveness within the solar industry. They also serve as a valuable backup for silicon photovoltaic, providing resilience to the overall energy infrastructure. However, the development of functional solar materials poses intricate multiobjective optimization challenges in a large multidimensional composition and parameter space, in some cases with millions of potential candidates to be explored. Solving it necessitates reproducible, user-independent laboratory work and intelligent preselection of innovative experimental methods.Materials acceleration platforms (MAPs) seamlessly integrate robotic materials synthesis and characterization with AI-driven data analysis and experimental design, positioning them as enabling technologies for the discovery and exploration of new materials. They are proposed to revolutionize materials development away from the Edisonian trial-and-error approaches to ultrashort cycles of experiments with exceptional precision, generating a reliable and highly qualitative data situation that allows training machine learning algorithms with predictive power. MAPs are designed to assist the researcher in multidimensional aspects of materials discovery, such as material synthesis, precursor preparation, sample processing and characterization, and data analysis, and are drawing escalating attention in the field of energy materials. Device acceleration platforms (DAPs), however, are designed to optimize functional films and layer stacks. Unlike MAPs, which focus on material discovery, a central aspect of DAPs is the identification and refinement of ideal processing conditions for a predetermined set of materials. Such platforms prove especially invaluable when dealing with "disordered semiconductors," which depend heavily on the processing parameters that ultimately define the functional properties and functionality of thin film layers. By facilitating the fine-tuning of processing conditions, DAPs contribute significantly to the advancement and optimization of disordered semiconductor devices, such as emerging photovoltaics.In this Account, we review the recent advancements made by our group in automated and autonomous laboratories for advanced material discovery and device optimization with a strong focus on emerging photovoltaics, such as solution-processing perovskite solar cells and organic photovoltaics. We first introduce two MAPs and two DAPs developed in-house: a microwave-assisted high-throughput synthesis platform for the discovery of organic interface materials, a multipurpose robot-based pipetting platform for the synthesis of new semiconductors and the characterization of thin film semiconductor composites, the SPINBOT system, which is a spin-coating DAP with the potential to optimize complex device architectures, and finally, AMANDA, a fully integrated and autonomously operating DAP. Notably, we underscore the utilization of a robot-based high-throughput experimentation technique to address the common optimization challenges encountered in extensive multidimensional composition and parameter spaces pertaining to organic and perovskite photovoltaics materials. Finally, we briefly propose a holistic concept and technology, a self-driven autonomous material and device acceleration platform (AMADAP) laboratory, for autonomous functional solar materials discovery and development. We hope to discover how AMADAP can be further strengthened and universalized with advancing development of hardware and software infrastructures in the future.

Acoustic 2-D Full-Waveform Inversion with Non-Balanced Finite differences and Adaptive Weight Decay Methods

Full waveform inversion (FWI) is a powerful seismic imaging technique that aims to achieve high-resolution subsurface models by iteratively comparing recorded seismic waveforms with modeled waveforms. The success of FWI relies on the precision of forward modeling algorithms and the effectiveness of optimization strategies. In this study, we propose a FWI methodology that leverages non-balanced staggered-grid finite differences schemes (NBFD) in the forward modeling process. By employing operators with spatial differencing of varying lengths, these schemes enhance the efficiency of forward modeling.For the optimization process, we employ adaptive gradient-based optimization with weight decay, a widely-used approach in machine learning and deep learning algorithms for model training. These methods demonstrate efficient optimization processing for large data volumes with high precision. The objective of adaptive gradient-based optimization is to dynamically adjust the update rate model parameters during the iterative process. This facilitates improved convergence speed, enhances stability, and overcomes challenges associated with traditional fixed-rate adjustment methods.To ensure accurate inversion results and avoid local minima in the cost function minimization, we implement a Multiscaling inversion strategy within the workflow. The obtained results are compared with those achieved through conventional FWI using standard staggered finite differences and various optimization methods. Our findings demonstrate that employing non-balanced staggered-grid finite differences scheme leads to significant improvements compared to the previously reported results in the literature.

Language model enhanced surface chloride concentration determination for concrete within splash environment based on limited field records

Chloride ion is severely harmful to reinforced concrete (RC) structures in marine environments. For maintaining the durability and safety of the designed RC structures, the determination of chloride ion concentration on concrete surfaces is critical. Currently, surface chloride ion concentration can be determined using empirical formulas and machine learning (ML) approaches. However, these approaches only rely on the numerical information within field records, disregarding valuable semantic and background information in the records, leading to low accuracy. Meanwhile, in splash environments, it presents a significant challenge to obtain chloride ion concentration records due to the complex environment and high costs involved. Therefore, based on limited field records of surface ion concentrations in splash environments, and utilizing a state-of-the-art language model (LM), this study proposes an LM-based information generation (LMIG) model to improve the accuracy of determination of surface chloride concentrations on RC structures. This paper utilizes the numerical and semantic information in 70 sets of field records to fine-tune the LMIG model and generates 200 sets of high-quality records. These records are then used to train ML algorithms for predicting chloride ion concentrations on concrete surfaces. After conducting comparative research, it was found that incorporating records generated by the LMIG model significantly enhances the accuracy of the ML algorithm. Specifically, the predictive accuracy using the random forest algorithm increased by 33.1%. Furthermore, this paper also conducts a comparative study on the LMIG model and the generative adversarial network (GAN)-assisted data-driven method. The results demonstrate that integrating semantic and numerical information into the LMIG model shows significant advantages in enhancing ML algorithms’ accuracy.

Classification of substances by health hazard using deep neural networks and molecular electron densities

In this paper we present a method that allows leveraging 3D electron density information to train a deep neural network pipeline to segment regions of high, medium and low electronegativity and classify substances as health hazardous or non-hazardous. We show that this can be used for use-cases such as cosmetics and food products. For this purpose, we first generate 3D electron density cubes using semiempirical molecular calculations for a custom European Chemicals Agency (ECHA) subset consisting of substances labelled as hazardous and non-hazardous for cosmetic usage. Together with their 3-class electronegativity maps we train a modified 3D-UNet with electron density cubes to segment reactive sites in molecules and classify substances with an accuracy of 78.1%. We perform the same process on a custom food dataset (CompFood) consisting of hazardous and non-hazardous substances compiled from European Food Safety Authority (EFSA) OpenFoodTox, Food and Drug Administration (FDA) Generally Recognized as Safe (GRAS) and FooDB datasets to achieve a classification accuracy of 64.1%. Our results show that 3D electron densities and particularly masked electron densities, calculated by taking a product of original electron densities and regions of high and low electronegativity can be used to classify molecules for different use-cases and thus serve not only to guide safe-by-design product development but also aid in regulatory decisions.Scientific contributionWe aim to contribute to the diverse 3D molecular representations used for training machine learning algorithms by showing that a deep learning network can be trained on 3D electron density representation of molecules. This approach has previously not been used to train machine learning models and it allows utilization of the true spatial domain of the molecule for prediction of properties such as their suitability for usage in cosmetics and food products and in future, to other molecular properties. The data and code used for training is accessible at https://github.com/s-singh-ivv/eDen-Substances.

Analysis of Data Generation and Preparation for Porosity Prediction in Cold Spray using Machine Learning

AbstractCold spray is an additive manufacturing and coating process in which powder particles are accelerated to supersonic speeds without melting them and then deposit on a surface to form a layer of a coating. Process parameters and materials affect the characteristics of manufactured parts and therefore must be chosen with care. Machine learning (ML) techniques have been specifically applied in additive manufacturing for tasks such as predicting and characterizing porosity. Machine learning algorithms can learn how a variation in the input spray parameters affects annotated output data, such as experimentally measured part properties. In this work, a dataset was developed from experiments reported in published academic papers, to train ML algorithms for the porosity prediction of cold spray manufactured parts. Data cleaning steps, such as null value replacement and categorical feature handling, were applied to prepare the dataset for the training of different ML models. The dataset was split into training and testing portions, and floating feature selection and hyperparameter optimization were performed using parts of the training set. A final evaluation of all trained models, using the test portion of the dataset, showed that a prediction accuracy with an average deviation of 0-2% porosity of the predicted values compared to the true values can be achieved. Graphical Abstract

Aircraft Engine Remaining Useful Life Prediction using neural networks and real-life engine operational data

Aircraft Engine Remaining Useful Life is a key factor which strongly affects flight operations safety and flight operators business decisions. In the article authors decided to present the concept of engine remaining useful life prediction. Proposed method was created as a result of the analysis of the real turbofan engine operational data collected for a few years which was used as an input data for the deep neural network, in order to train, validate and test machine learning algorithms. Two architectures of deep neural networks were created: multilayered deep convolutional neural networks and a long short-term memory network with regression output. Both neural networks were trained, validated and tested on the same engine data and with a various network training options. Results were compared with the neural network performance metrics and figures presenting network prediction convergence. To present how the real-life engine dataset differs the results from the simulated data, both datasets were validated on the same neural network architectures. The main purpose of this article was to present the idea and method of how the artificial neural networks could be used to predict aircraft remaining useful life indicator on the real-life engine operational data not the simulated one.

A finite element-based machine learning framework to predict the mechanical behavior of the pelvic floor muscles during childbirth

The medical community has been focusing on gaining a deeper understanding of birth trauma, which affects millions of women worldwide. Maternal lesions can be challenging to diagnose and expensive to examine. To better comprehend the mechanism of injuries occurring in the pelvic floor muscles (PFM), biomechanical simulations can be a valuable tool. However, utilizing the finite element method (FEM) to conduct simulations can be a time-consuming process. To overcome this issue, the present study aims to develop a machine learning (ML) framework to predict stresses on the PFM during childbirth by training ML algorithms on FEM simulation data. To generate the dataset for the ML algorithm’s training, childbirth simulations were performed using different material properties to characterize the PFM. Four ML algorithms were employed, namely Random Forest (RF), Extreme Gradient Boosting (XGBT), Support Vector Regression (SVR), and Artificial Neural Networks (ANN), considering two scenarios: (1) stress prediction for the maximum stretch level of the muscle, and (2) for multiple levels of fetal descent. Results showed that the ANN performed best in the former, with a mean absolute error (MAE) of 0.191 MPa. In the latter, XGBT provided lower errors for 20 and 35 mm of fetal descent, with MAE values of 0.002 and 0.028 MPa, respectively. Nevertheless, the ANN yielded better predictions for 50 and 65 mm, with MAE values of 0.214 and 0.187 MPa, respectively. The present work represents the first attempt to use FEM-based ML algorithms with childbirth simulations to obtain near-real-time predictions in routine clinical procedures.

Non-Invasive Heart Failure Evaluation Using Machine Learning Algorithms.

Heart failure is a prevalent cardiovascular condition with significant health implications, necessitating effective diagnostic strategies for timely intervention. This study explores the potential of continuous monitoring of non-invasive signals, specifically integrating photoplethysmogram (PPG) and electrocardiogram (ECG), for enhancing early detection and diagnosis of heart failure. Leveraging a dataset from the MIMIC-III database, encompassing 682 heart failure patients and 954 controls, our approach focuses on continuous, non-invasive monitoring. Key features, including the QRS interval, RR interval, augmentation index, heart rate, systolic pressure, diastolic pressure, and peak-to-peak amplitude, were carefully selected for their clinical relevance and ability to capture cardiovascular dynamics. This feature selection not only highlighted important physiological indicators but also helped reduce computational complexity and the risk of overfitting in machine learning models. The use of these features in training machine learning algorithms led to a model with impressive accuracy (98%), sensitivity (97.60%), specificity (96.90%), and precision (97.20%). Our integrated approach, combining PPG and ECG signals, demonstrates superior performance compared to single-signal strategies, emphasizing its potential in early and precise heart failure diagnosis. The study also highlights the importance of continuous monitoring with wearable technology, suggesting a significant stride forward in non-invasive cardiovascular health assessment. The proposed approach holds promise for implementation in hardware systems to enable continuous monitoring, aiding in early detection and prevention of critical health conditions.

Assessing SOFA score trajectories in sepsis using machine learning: A pragmatic approach to improve the accuracy of mortality prediction.

An increasing amount of longitudinal health data is available on critically ill septic patients in the age of digital medicine, including daily sequential organ failure assessment (SOFA) score measurements. Thus, the assessment in sepsis focuses increasingly on the evaluation of the individual disease's trajectory. Machine learning (ML) algorithms may provide a promising approach here to improve the evaluation of daily SOFA score dynamics. We tested whether ML algorithms can outperform the conventional ΔSOFA score regarding the accuracy of 30-day mortality prediction. We used the multicentric SepsisDataNet.NRW study cohort that prospectively enrolled 252 sepsis patients between 03/2018 and 09/2019 for training ML algorithms, i.e. support vector machine (SVM) with polynomial kernel and artificial neural network (aNN). We used the Amsterdam UMC database covering 1,790 sepsis patients for external and independent validation. Both SVM (AUC 0.84; 95% CI: 0.71-0.96) and aNN (AUC 0.82; 95% CI: 0.69-0.95) assessing the SOFA scores of the first seven days led to a more accurate prognosis of 30-day mortality compared to the ΔSOFA score between day 1 and 7 (AUC 0.73; 95% CI: 0.65-0.80; p = 0.02 and p = 0.05, respectively). These differences were even more prominent the shorter the time interval considered. Using the SOFA scores of day 1 to 3 SVM (AUC 0.82; 95% CI: 0.68 0.95) and aNN (AUC 0.80; 95% CI: 0.660.93) led to a more accurate prognosis of 30-day mortality compared to the ΔSOFA score (AUC 0.66; 95% CI: 0.58-0.74; p < 0.01 and p < 0.01, respectively). Strikingly, all these findings could be confirmed in the independent external validation cohort. The ML-based algorithms using daily SOFA scores markedly improved the accuracy of mortality compared to the conventional ΔSOFA score. Therefore, this approach could provide a promising and automated approach to assess the individual disease trajectory in sepsis. These findings reflect the potential of incorporating ML algorithms as robust and generalizable support tools on intensive care units.

Quantifying Seagrass Density Using Sentinel-2 Data and Machine Learning

Seagrasses, rooted aquatic plants growing completely underwater, are extremely important for the coastal ecosystem. They are an important component of the total carbon burial in the ocean, they provide food, shelter, and nursery to many aquatic organisms in coastal ecosystems, and they improve water quality. Due to human activity, seagrass coverage has been rapidly declining, and there is an urgent need to monitor seagrasses consistently. Seagrass coverage has been closely monitored in the Chesapeake Bay since 1970 using air photos and ground samples. These efforts are costly and time-consuming. Many studies have used remote sensing data to identify seagrass bed outlines, but few have mapped seagrass bed density. This study used Sentinel-2 satellite data and machine learning in Google Earth Engine and the Chesapeake Bay Program field data to map seagrass density. We used seagrass density data from the Chincoteague and Sinepuxent Bay to train machine learning algorithms and evaluate their accuracies. Out of the four machine learning models tested (Naive Bayes (NB), Classification and Regression Trees (CART), Support Vector Machine (SVM), and Random Forest (RF)), the RF model outperformed the other three models with overall accuracies of 0.874 and Kappa coefficients of 0.777. The SVM and CART models performed similarly and NB performed the poorest. We tested two different approaches to assess the models’ accuracy. When we used all the available ground samples to train the models, whereby our analysis showed that model performance was associated with seagrass density class, and that higher seagrass density classes had better consumer accuracy, producer accuracy, and F1 scores. However, the association of model performance with seagrass density class disappeared when using the same training data size for each class. Very sparse and dense seagrass classes had replacedhigherbetter accuracies than the sparse and moderate seagrass density classes. This finding suggests that training data impacts machine learning model performance. The uneven training data size for different classes can result in biased assessment results. Selecting proper training data and machine learning models are equally important when using machine learning and remote sensing data to map seagrass density. In summary, this study demonstrates the potential to map seagrass density using satellite data.

Impact of Enterprises Relationship on ML’s Ability to Forecast SME’s Credit Risk in SCF: A Research Based on Complex Network

ABSTRACT In this study, we investigate the impact of introducing enterprise relationship to train machine learning (ML) algorithms on improving its ability of forecasting small and medium-sized enterprises’ (SMEs) credit risk in supply chain finance (SCF). First, we incorporate attributes of SME node in the CEs-SMEs network, a bipartite network constructed with transaction data in the Chinese automobile industry, into financial indicators to train ML algorithms for improving forecasting performance. Second, we employ the weighted one-mode projection approach to extract a projected-SMEs network that reveals the relationship among SMEs from the CEs-SMEs network and examine whether the ML algorithms’ forecasting performances is improved when considering attributes in the projected-SMEs network. Third, we analyze the correlation between SME’s credit risk and individual network attribute to specifically understand the impact of enterprise relationship on credit risk. Overall, the empirical results indicate that the forecasting performance is obviously better when enterprise transaction relationship is considered, and it is further enhanced after we apply weighted one-mode projection approach. Meanwhile, the variable importance analysis and the binomial logistic regression demonstrate how significantly each network attribute is correlated with credit risk, and the discussion on partial dependence plot shows that SMEs with large degree are non-risky generally.

BioDiffusion: A Versatile Diffusion Model for Biomedical Signal Synthesis.

Machine learning tasks involving biomedical signals frequently grapple with issues such as limited data availability, imbalanced datasets, labeling complexities, and the interference of measurement noise. These challenges often hinder the optimal training of machine learning algorithms. Addressing these concerns, we introduce BioDiffusion, a diffusion-based probabilistic model optimized for the synthesis of multivariate biomedical signals. BioDiffusion demonstrates excellence in producing high-fidelity, non-stationary, multivariate signals for a range of tasks including unconditional, label-conditional, and signal-conditional generation. Leveraging these synthesized signals offers a notable solution to the aforementioned challenges. Our research encompasses both qualitative and quantitative assessments of the synthesized data quality, underscoring its capacity to bolster accuracy in machine learning tasks tied to biomedical signals. Furthermore, when juxtaposed with current leading time-series generative models, empirical evidence suggests that BioDiffusion outperforms them in biomedical signal generation quality.

Abstract 4906: Biopsy free virtual histology to discriminate benign from malignant squamous neoplasms

Abstract Reflectance confocal microscopy (RCM) is a noninvasive optical imaging modality that allows for cellular-level resolution, in vivo images of skin without performing a traditional skin biopsy. RCM image interpretation currently requires specialized training to interpret the grayscale output images that are difficult to correlate with tissue pathology. Here, we use a deep learning-based framework that uses a convolutional neural network to transform grayscale output images into virtually-stained hematoxylin and eosin (H&amp;E)-like images allowing for the visualization of various skin layers, including the epidermis, dermal-epidermal junction, and superficial dermis layers. To train the deep-learning framework, a stack of a minimum of 7 time-lapsed, successive RCM images of excised tissue were obtained from epidermis to dermis 1.52 microns apart to a depth of 60.96 microns using the Vivascope 3000. The tissue was embedded in agarose tissue and a curette was used to create a tunnel through which drops of 50% acetic acid was used to stain cell nuclei. These acetic acid-stained images were used as “ground truth” to train a deep convolutional neural network using a conditional generative adversarial network (GAN)-based machine learning algorithm to digitally convert the images into GAN-based H&amp;E-stained digital images. We used the already trained machine learning algorithm and retrained the algorithm with new samples to include squamous neoplasms. Through further training and refinement of the algorithm, high-resolution, histological quality images can be obtained to aid in earlier diagnosis and treatment of cutaneous neoplasms. The overall goal of obtaining biopsy-free virtual histology images with this technology can be used to provide real-time outputs of virtually-stained H&amp;E skin lesions, thus decreasing the need for invasive diagnostic procedures and enabling greater uptake of the technology by the medical community. Citation Format: Nelly Kokikian, Rachel Wahhab, Jingxi Li, Stephanie Martin, David P. Beynet, Aydogan Ozcan, Phillip Scumpia. Biopsy free virtual histology to discriminate benign from malignant squamous neoplasms [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 4906.

ENHANCING SUBJECTIVE ANSWER EVALUATION THROUGH MACHINE LEARNING AND NATURAL LANGUAGE PROCESSING

This research introduces a pioneering framework that harnesses machine learning and natural language processing (NLP) to revolutionize the evaluation of subjective answers in educational contexts. Traditional methods of assessing essays and open-ended responses have been characterized by their labour-intensive nature and subjectivity. Our approach streamlines this process by employing NLP techniques for preprocessing, tokenization, and advanced feature extraction, followed by training machine learning algorithms on diverse datasets of annotated answers. The result is a system capable of providing automated scores and feedback that closely align with human evaluators' judgments, demonstrating effectiveness and reliability across a spectrum of educational domains. Importantly, this automation not only enhances scalability and consistency but also lightens the workload on educators, allowing them to focus on more nuanced aspects of teaching. Beyond its technical contributions, our research addresses ethical considerations and challenges associated with the deployment of automated evaluation systems in educational settings. This comprehensive exploration encompasses concerns related to bias, transparency, and the overall impact on the learning experience. By navigating these ethical dimensions, our study not only advances the technological aspects of automated evaluation but also underscores the importance of responsible implementation within the educational landscape. This dual emphasis on technical innovation and ethical considerations positions our framework as a promising solution for achieving efficient and objective subjective answer assessment in educational contexts. Keywords: Machine learning, NLP, Subjective answer assessment, automatic scoring, feature extraction, consistency, feedback, teaching work load reduction, transparent evaluation

Abstract 4970: Multi-modal machine learning approaches for predicting cancer type and Gleason grade leveraging public TCGA data

Abstract Introduction: To better understand the complex and challenging nature of diseases such as cancer and for improved diagnosis, it may require the combination of multiple data modalities, such as histopathological images and omics data such as RNA-seq. By integrating these heterogeneous but complementary data, a multimodal approach unites both worlds and could achieve better synergistic results compared to using a single modality. The growing availability of large datasets such as The Cancer Genome Atlas (TCGA) with more than 10000 patients made it possible to combine different modalities to train machine learning algorithms which offers great potential to address challenging cancer related research. In this proof of concept initiative we use machine learning approaches within an open-source framework in order to leverage the potential of multimodality (Histopathology Whole Slide Images (WSI) and Genomics/RNA-seq) to build predictive AI models for cancer type and prostate Gleason score, and provide a potential to develop a quality control step. Method: We used matched WSI and RNA-Seq profiles from TCGA, including 11093 samples and 30 cancer types to develop a pancancer classification model using both modalities. For prostate Gleason score prediction 401 patients were available. Both datasets were split into a train (70%) and test (30%) components. We used a late fusion approach where we combined the RNA-seq model (linear SVM) with the WSI model (Resnet18) by multiplying the probability scores of each single-modality model. Model performance was measured with the F1 metric. Results: For cancer type prediction, the multimodality model achieved an F1 score of 0.95 on the test set. About 40% of the cancer types benefited from a synergistic effect by combining the two modalities. Cancer types and percent increase in F1 scores, respectively, that benefit most by combining modalities are: Cervical squamous cell carcinoma and endocervical adenocarcinoma (4.23%), Cholangiosarcoma (6.66%) and Uterine carcinosarcoma (4%). Interestingly, in other cancer types the combination did not result in improved predictive scores compared to a single modality model, e.g. in Rectum adenocarcinoma, Sarcoma or Stomach adenocarcinoma. For Prostate cancer grading, Gleason score prediction of patterns 3/4/5, combined multi modality model earned 0.73 F1 outperforming the single modality models. Conclusion: By combining histopathology imaging and omics modalities we demonstrated synergistic effects in predictive power for both cancer-related research questions. We show improved predictive performance in 40% of the classified cancer types by taking both modalities. Imaging or omics modalities alone can be sufficient in some cases and their strengths are very problem-specific. Citation Format: Christian Wohlfart, Eldad Klaiman, Jacub Witkowski, Michael King, Jacob Gildenblat, Ofir Etz-Hadar, Mohammad Ashtari, Antoaneta Vladimirova. Multi-modal machine learning approaches for predicting cancer type and Gleason grade leveraging public TCGA data [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 4970.

Abstract 6087: Cytomics-on-chip and AI-driven predictive analysis platform for early detection of epithelial cancers

Abstract Cancer is the 2nd leading cause of death (over 605,000 people) in the US, at an expense of over $200B, with 1 in 3 people projected to have cancer during their lifetime per CDC. Despite the significant impact of early detection and screening on prognosis, only some cancers are diagnosed at an early stage. Carcinomas, comprising &amp;gt;80% of cancer incidence, allow ease in cytology sample access chairside, due to the lesions’ epithelial presentation. This presents a unique opportunity for early detection and screening in epithelial cancers. In low-resource healthcare settings, from clinical examination to the long, tedious and expensive diagnostic journey for cancers &amp; pre-cancerous lesions, can lead to missed, delayed or over diagnosis scenarios. This affects treatment initiation and potentially outcome. To facilitate early intervention, there is compelling need to develop accurate and effective minimally invasive screening platforms. Recent advances in the -omics disciplines, microfluidics and AI tools are starting to reveal promising signatures of early disease detection, with potential to drastically improve screening and diagnostic systems. We are developing a novel application of the cytomics-on-chip platform, for enabling chairside, quantitative screening of suspicious epithelial lesions. The biosensor module involves 1. a single use, cytomics platform employing a cartridge with cellular array and high specificity biomarker reagents, that allows single cell molecular imaging to be completed in a portable analyzer. 2. a microfluidics module that allows cytomorphometric measurements to be completed. 3. The results generated are utilized to train machine learning algorithms to detect cyto-signatures and provide an intuitive result that may be utilized by health care practitioners in clinical-decision making. The first cell-based point-of-care oncology tool has recently been validated with high accuracy (99.3%), sensitivity and specificity, in a multi-site prospective clinical study. Here we demonstrate a pilot study towards development of a smart single cell cytomics-on-chip platform for prompt cytomorphometric and biomarker characterization towards diagnosis of urothelial, anal and cervical cancers, and dysplasia lesions, utilizing brush/pap and fresh urine samples. This has potential for continuous quantitative indexing, for disease categorization. As cancers become more pervasive, improved early detection/screening methods that are accurate, cost effective, easy to implement during routine clinical practice, and providing minimal discomfort to the patient, are urgently needed, improving confidence in clinicians’ decisions. Citation Format: Kritika Srinivasan Rajsri, Michael P. McRae, Nicolaos J. Christodoulides, Khaled Algashaamy, Monica T. Garcia-Buitrago, Fei Chen, Fang-Ming Deng, Jennifer S. Smith, John T. McDevitt. Cytomics-on-chip and AI-driven predictive analysis platform for early detection of epithelial cancers [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 6087.