Machine Learning-based Classification Models Research Articles

Development of Machine Learning Model for Breast Cancer Prediction from Ultrasound Images

In the past decade, the revolution in information and computing technology has transformed approaches to breast cancer detection and treatment, with Machine Learning technologies offering significant potential in health data analysis. However, the development of accurate and reliable predictive models is faced with the challenges of data heterogeneity and complexity. This research proposes the development and validation of Machine Learning-based classification models using Support Vector Machine and Principal Component Analysis to address these issues, targeting improved accuracy in the early detection of breast cancer. The methodology applied involved the use of a breast cancer dataset from Kaggle, with data analysis conducted through inductive methods to identify relevant patterns. The combination of Support Vector Machine and Principal component Analysis achieved 89% accuracy in medical image classification, proving its efficacy in breast cancer diagnostics and providing a more reliable model for early detection. The implications of these findings are significant, both theoretically and practically, for the fields of Machine Learning and Breast Cancer, expanding the understanding of the applications of advanced data processing techniques. Although this study faces limitations in the variability of the dataset's patient characteristics, the results offer a basis for further development in diagnostic technology while recommending the integration of Deep Learning and Big Data analysis as a direction for future research.

Machine Learning Based Classification Model for Prediction of Bank Loan Approval

Abstract: In this paper we creating and testing a smart loan prediction system using a special type of classifier called gradient boosting. Our methodology is meticulously laid out, covering a spectrum of essential stages including data pre-processing methodologies, in-depth exploratory analysis, and rigorous model training leveraging a dataset meticulously curated from Kaggle. Our model, with 300 estimators, a learning rate of 0.05, and a maximum of six features, achieves an astounding 98.03% accuracy. We utilize the pickle module to serialize our trained model, consequently deploying it as a web application. This application, which is powered by React for the front end and Flask for the back end, offers users a simple interface for entering loan application details and receiving real-time forecasts on loan acceptance or denial. Our research shows that gradient boosting classified the well in predicting the loans to the users.

Discovery of an independent poor-prognosis subtype associated with tertiary lymphoid structures in breast cancer.

Tertiary lymphoid structures (TLSs) are ectopic lymphoid formations that arise in non-lymphoid tissues due to chronic inflammation. The pivotal function of TLSs in regulating tumor invasion and metastasis has been established across several cancers, such as lung cancer, liver cancer, and melanoma, with a positive correlation between increased TLS presence and improved prognosis. Nevertheless, the current research about the clinical significance of TLSs in breast cancer remains limited. In our investigation, we discovered TLS-critical genes that may impact the prognosis of breast cancer patients, and categorized breast cancer into three distinct subtypes based on critical gene expression profiles, each exhibiting substantial differences in prognosis (p = 0.0046, log-rank test), with Cluster 1 having the best prognosis, followed by Cluster 2, and Cluster 3 having the worst prognosis. We explored the impact of the heterogeneity of these subtypes on patient prognosis, the differences in the molecular mechanism, and their responses to drug therapy and immunotherapy. In addition, we designed a machine learning-based classification model, unveiling highly consistent prognostic distinctions in several externally independent cohorts. A notable marker gene CXCL13 was identified in Cluster 3, potentially pivotal in enhancing patient prognosis. At the single-cell resolution, we delved into the adverse prognosis of Cluster 3, observing an enhanced interaction between fibroblasts, myeloid cells, and basal cells, influencing patient prognosis. Furthermore, we identified several significantly upregulated genes (CD46, JAG1, IL6, and IL6R) that may positively correlate with cancer cells' survival and invasive capabilities in this subtype. Our study is a robust foundation for precision medicine and personalized therapy, presenting a novel perspective for the contemporary classification of breast cancer.

Community-enhanced Link Prediction in Dynamic Networks

The growing popularity of online social networks is quite evident nowadays and provides an opportunity to allow researchers in finding solutions for various practical applications. Link prediction is the technique of understanding network structure and identifying missing and future links in social networks. One of the well-known classes of methods in link prediction is a similarity-based method, which uses local and global topological information of the network to predict missing links. Some methods also exist based on quasi-local features to achieve a trade-off between local and global information on static networks. These quasi-local similarity-based methods are not best suited for considering community information in dynamic networks, failing to balance accuracy and efficiency. Therefore, a community-enhanced framework is presented in this article to predict missing links on dynamic social networks. First, a link prediction framework is presented to predict missing links using parameterized influence regions of nodes and their contribution in community partitions. Then, a unique feature set is generated using local, global, and quasi-local similarity-based as well as community information-based features. This feature set is further optimized using scoring-based feature selection methods to select only the most relevant features. Finally, four machine learning-based classification models are used for link prediction. The experiments are performed on six well-known dynamic networks and three performance metrics, and the results demonstrate that the proposed method outperforms the state-of-the-art methods.

A machine learning-based classification model to support university students with dyslexia with personalized tools and strategies

Dyslexia is a specific learning disorder that causes issues related to reading, which affects around 10% of the worldwide population. This can compromise comprehension and memorization skills, and result in anxiety and lack of self-esteem, if no support is provided. Moreover, this support should be highly personalized, to be actually helpful. In this paper, a model to classify the most useful methodologies to support students with dyslexia has been created, with a focus on university alumni. The prediction algorithm is based on supervised machine learning techniques; starting from the issues that dyslexic students experience during their career, it is capable of suggesting customized support digital tools and learning strategies for each of them. The algorithm was trained and tested on data acquired through a self-evaluation questionnaire, which was designed and then spread to more than 1200 university students. It allowed 17 useful tools and 22 useful strategies to be detected. The results of the testing showed an average prediction accuracy higher than 90%, which rises to 94% by renouncing to guess the less-predictable 8 tools/strategies. In the light of this, it is possible to state that the implemented algorithm can achieve the set goal and, thus, reduce the gap between dyslexic and non-dyslexic students. This achievement paves the way for a new modality of facing the problem of dyslexia by university institutions, which aims at modifying teaching activities toward students’ needs, instead of simply reducing their study load or duties. This complies with the definition and the aims of inclusivity.

Stress phenotyping analysis leveraging autofluorescence image sequences with machine learning.

Autofluorescence-based imaging has the potential to non-destructively characterize the biochemical and physiological properties of plants regulated by genotypes using optical properties of the tissue. A comparative study of stress tolerant and stress susceptible genotypes of Brassica rapa with respect to newly introduced stress-based phenotypes using machine learning techniques will contribute to the significant advancement of autofluorescence-based plant phenotyping research. Autofluorescence spectral images have been used to design a stress detection classifier with two classes, stressed and non-stressed, using machine learning algorithms. The benchmark dataset consisted of time-series image sequences from three Brassica rapa genotypes (CC, R500, and VT), extreme in their morphological and physiological traits captured at the high-throughput plant phenotyping facility at the University of Nebraska-Lincoln, USA. We developed a set of machine learning-based classification models to detect the percentage of stressed tissue derived from plant images and identified the best classifier. From the analysis of the autofluorescence images, two novel stress-based image phenotypes were computed to determine the temporal variation in stressed tissue under progressive drought across different genotypes, i.e., the average percentage stress and the moving average percentage stress. The study demonstrated that both the computed phenotypes consistently discriminated against stressed versus non-stressed tissue, with oilseed type (R500) being less prone to drought stress relative to the other two Brassica rapa genotypes (CC and VT). Autofluorescence signals from the 365/400 nm excitation/emission combination were able to segregate genotypic variation during a progressive drought treatment under a controlled greenhouse environment, allowing for the exploration of other meaningful phenotypes using autofluorescence image sequences with significance in the context of plant science.

Advancing device and network security for enhanced privacy

The rapid proliferation of IoT (Internet of Things) devices has ushered in an era of unprecedented connectivity and automation. The widespread adoption has also exposed vulnerabilities, necessitating robust security and privacy measures. This research presents a comprehensive study focused on enhancing IoT device and network security and privacy through empirical investigation and advanced machine learning techniques. Commencing with an exhaustive literature review, it was assessed, the evolving landscape of IoT security threats, solutions, and identified research gaps. Building upon the foundation, it was designed and rigorously evaluated a machine learning-based classification model tailored for IoT device security. Utilizing a meticulously crafted simulated dataset mirroring real-world IoT features, our model undergoes comprehensive performance evaluations. Metrics include accuracy, precision, recall, F1 score, and ROC analysis. Our findings reveal a nuanced performance profile, shedding light on the model's capability to accurately classify IoT devices as Secure or Vulnerable. Precision-recall trade-offs, emphasizing the need for a judicious balance to mitigate false positives and false negatives was investigated. The critical role of feature engineering and model refinement, pointing to areas for future research and optimization. This research contributes to the burgeoning field of IoT security by employing machine learning as a proactive tool for fortifying IoT device and network security. Our findings advocate for a strategic approach to secure IoT ecosystems, ensuring data integrity and privacy in the face of evolving threats. As IoT devices continue to proliferate across industries, this research serves as a foundation for innovative strategies and ongoing investigations to harness the full potential of secure IoT environments while addressing multifaceted challenges in the ever-evolving IoT landscape.

DeepGraFT: A novel semantic segmentation auxiliary ROI-based deep learning framework for effective fundus tessellation classification

Fundus tessellation (FT) is a prevalent clinical feature associated with myopia and has implications in the development of myopic maculopathy, which causes irreversible visual impairment. Accurate classification of FT in color fundus photo can help predict the disease progression and prognosis. However, the lack of precise detection and classification tools has created an unmet medical need, underscoring the importance of exploring the clinical utility of FT. Thus, to address this gap, we introduce an automatic FT grading system (called DeepGraFT) using classification-and-segmentation co-decision models by deep learning. ConvNeXt, utilizing transfer learning from pretrained ImageNet weights, was employed for the classification algorithm, aligning with a region of interest based on the ETDRS grading system to boost performance. A segmentation model was developed to detect FT exits, complementing the classification for improved grading accuracy. The training set of DeepGraFT was from our in-house cohort (MAGIC), and the validation sets consisted of the rest part of in-house cohort and an independent public cohort (UK Biobank). DeepGraFT demonstrated a high performance in the training stage and achieved an impressive accuracy in validation phase (in-house cohort: 86.85 %; public cohort: 81.50 %). Furthermore, our findings demonstrated that DeepGraFT surpasses machine learning-based classification models in FT classification, achieving a 5.57 % increase in accuracy. Ablation analysis revealed that the introduced modules significantly enhanced classification effectiveness and elevated accuracy from 79.85 % to 86.85 %. Further analysis using the results provided by DeepGraFT unveiled a significant negative association between FT and spherical equivalent (SE) in the UK Biobank cohort. In conclusion, DeepGraFT accentuates potential benefits of the deep learning model in automating the grading of FT and allows for potential utility as a clinical-decision support tool for predicting progression of pathological myopia.

Identifying primary aldosteronism patients who require adrenal venous sampling: a multi-center study

Adrenal venous sampling (AVS) is crucial for subtyping primary aldosteronism (PA) to explore the possibility of curing hypertension. Because AVS availability is limited, efforts have been made to develop strategies to bypass it. However, it has so far proven unsuccessful in applying clinical practice, partly due to heterogeneity and missing values of the cohorts. For this purpose, we retrospectively assessed 210 PA cases from three institutions where segment-selective AVS, which is more accurate and sensitive for detecting PA cases with surgical indications, was available. A machine learning-based classification model featuring a new cross-center domain adaptation capability was developed. The model identified 102 patients with PA who benefited from surgery in the present cohort. A new data imputation technique was used to address cross-center heterogeneity, making a common prediction model applicable across multiple cohorts. Logistic regression demonstrated higher accuracy than Random Forest and Deep Learning [(0.89, 0.86) vs. (0.84, 0.84), (0.82, 0.84) for surgical or medical indications in terms of f-score]. A derived integrated flowchart revealed that 35.2% of PA cases required AVS with 94.1% accuracy. The present model enabled us to reduce the burden of AVS on patients who would benefit the most.

An Automated Grading System Based on Topological Features for the Evaluation of Corneal Fluorescein Staining in Dry Eye Disease

Background: Corneal fluorescein staining is a key biomarker in evaluating dry eye disease. However, subjective scales of corneal fluorescein staining are lacking in consistency and increase the difficulties of an accurate diagnosis for clinicians. This study aimed to propose an automatic machine learning-based method for corneal fluorescein staining evaluation by utilizing prior information about the spatial connection and distribution of the staining region. Methods: We proposed an end-to-end automatic machine learning-based classification model that consists of staining region identification, feature signature construction, and machine learning-based classification, which fully scrutinizes the multiscale topological features together with conventional texture and morphological features. The proposed model was evaluated using retrospective data from Beijing Tongren Hospital. Two masked ophthalmologists scored images independently using the Sjögren’s International Collaborative Clinical Alliance Ocular Staining Score scale. Results: A total of 382 images were enrolled in the study. A signature with six topological features, two textural features, and two morphological features was constructed after feature extraction and selection. Support vector machines showed the best classification performance (accuracy: 82.67%, area under the curve: 96.59%) with the designed signature. Meanwhile, topological features contributed more to the classification, compared with other features. According to the distribution and correlation with features and scores, topological features performed better than others. Conclusions: An automatic machine learning-based method was advanced for corneal fluorescein staining evaluation. The topological features in presenting the spatial connectivity and distribution of staining regions are essential for an efficient corneal fluorescein staining evaluation. This result implies the clinical application of topological features in dry-eye diagnosis and therapeutic effect evaluation.

AIS-based operational phase identification using Progressive Ablation Feature Selection with machine learning for improving ship emission estimates

ABSTRACT The work status of ships’ engines and boilers has a significant impact on emission estimates, which are closely related to ships’ operational phases. To improve the accuracy of emission estimates, this study proposed a machine learning-based classification model for identifying operational phases. We proposed 12 operational phase relevance features by analyzing motion behavior-related and geospatial characteristics-related features from the Automatic Identification System (AIS) data from the two bulk carriers. The random forest (RF) model showed the best performance in identifying one of the bulk carrier’s operational phases among the five machine models, with the accuracy, F1score and Area Under Curve (AUC) of 96.66%, 93.34% and 99.93%, respectively. By adopting the Progressive Ablation Feature Selection (PAFS) method with RF, the number of features was reduced from 12 to 8, and the accuracy (96.38%), F1score (92.70%), and AUC (98.81%) were almost same with that obtained from all 12 features. Additionally, the effectiveness of the RF model was validated on the other bulk carriers. Compared with the traditional algorithms, the RF model showed better performance in ship operational phase identification and improved the average accuracy of NOx emission estimation for the main engine and auxiliary engine by 57.83% and 93.89%, respectively, under different operational phases. These results provide the basis for port traffic management and ship emission control. Implications: A new ship operational phase identification approach was proposed in this study. If the proposed approach is adopted by International Maritime Organization, it will improve the accuracy of ship emission estimates and bring new insights into global shipping greenhouse gas (GHG) emissions and their impact on global change. The port authorities could benefit from the proposed approach, which can be extended to ship types with similar behavior to bulk carriers, such as containers and general cargoes. This can reveal patterns of ship behavior in specific areas, which helps to identify potential collision risks, channel blockages, and other safety issues and take appropriate management measures to ensure the safe operation of the port. The proposed approach can help shipping companies to accurately estimate the GHG emissions of their fleets and to accurately predict carbon tax costs. Base on that, carbon emissions and carbon tax burden can be reduced by adopting corresponding management control measures.

Leveraging genetic data for predicting consumer choices of alcoholic products

PurposeFood choices profoundly affect one's dietary, nutritional and health outcomes. Using alcoholic beverages as a case study, the authors assess the potential of genetic data in predicting consumers' food choices combined with conventional socio-demographic data.Design/methodology/approachA discrete choice experiment was conducted to elicit the underlying preferences of 484 participants from seven provinces in China. By linking three types of data (—data from the choice experiment, socio-demographic information and individual genotyping data) of the participants, the authors employed four machine learning-based classification (MLC) models to assess the performance of genetic information in predicting individuals' food choices.FindingsThe authors found that the XGBoost algorithm incorporating both genetic and socio-demographic data achieves the highest prediction accuracy (77.36%), significantly outperforming those using only socio-demographic data (permutation test p-value = 0.033). Polygenic scores of several behavioral traits (e.g. depression and height) and genetic variants associated with bitter taste perceptions (e.g. TAS2R5 rs2227264 and TAS2R38 rs713598) offer contributions comparable to that of standard socio-demographic factors (e.g. gender, age and income).Originality/valueThis study is among the first in the economic literature to empirically demonstrate genetic factors' important role in predicting consumer behavior. The findings contribute fresh insights to the realm of random utility theory and warrant further consumer behavior studies integrating genetic data to facilitate developments in precision nutrition and precision marketing.

Machine learning-based classification models for non-covalent Bruton's tyrosine kinase inhibitors: predictive ability and interpretability.

In this study, we built classification models using machine learning techniques to predict the bioactivity of non-covalent inhibitors of Bruton's tyrosine kinase (BTK) and to provide interpretable and transparent explanations for these predictions. To achieve this, we gathered data on BTK inhibitors from the Reaxys and ChEMBL databases, removing compounds with covalent bonds and duplicates to obtain a dataset of 3895 inhibitors of non-covalent. These inhibitors were characterized using MACCS fingerprints and Morgan fingerprints, and four traditional machine learning algorithms (decision trees (DT), random forests (RF), support vector machines (SVM), and extreme gradient boosting (XGBoost)) were used to build 16 classification models. In addition, four deep learning models were developed using deep neural networks (DNN). The best model, Model D_4, which was built using XGBoost and MACCS fingerprints, achieved an accuracy of 94.1% and a Matthews correlation coefficient (MCC) of 0.75 on the test set. To provide interpretable explanations, we employed the SHAP method to decompose the predicted values into the contributions of each feature. We also used K-means dimensionality reduction and hierarchical clustering to visualize the clustering effects of molecular structures of the inhibitors. The results of this study were validated using crystal structures, and we found that the interaction between the BTK amino acid residue and the important features of clustered scaffold was consistent with the known properties of the complex crystal structures. Overall, our models demonstrated high predictive ability and a qualitative model can be converted to a quantitative model to some extent by SHAP, making them valuable for guiding the design of new BTK inhibitors with desired activity.

Spontaneous physical functional recovery after hospitalization for COVID-19: insights from a 1 month follow-up and a model to predict poor trajectory.

Long COVID syndrome has emerged as a new global healthcare challenge, with impaired physical performance being a prominent debilitating factor. Cardiopulmonary rehabilitation is a mainstay of management of symptomatic post-COVID patients, and optimization of candidate selection might allow for more effective use of available resources. In order to study the natural dynamics and to identify predictors of physical functional recovery following hospitalization for COVID-19, 6 min walk test was performed pre-discharge in 176 patients (40% hypertensive, 53% female, mean age 53.2 ± 13.5 years) with re-evaluation at 1 month. Six min walk distance and the reached percent of predicted distance (6MWD%) were suboptimal at both visits-396 ± 71 m (68.7 ± 12.4%) pre-discharge and 466 ± 65 m (81.8 ± 13.6%) at 1 month. Associated changes included significant oxygen desaturation (2.9 ± 2.5 and 2.3 ± 2.2%, respectively) and insufficient increment of heart rate during the test (24.9 ± 17.5 and 28.2 ± 12.0 bpm) that resulted in low reached percent of individual maximum heart rate (61.1 ± 8.1 and 64.3 ± 8.2%). Automatic clusterization of the study cohort by the 6MWD% changes has allowed to identify the subgroup of patients with poor "low base-low increment" trajectory of spontaneous post-discharge recovery that were characterized by younger age (38.2 ± 11.0 vs. 54.9 ± 12.1, p < 0.001) but more extensive pulmonary involvement by CT (43.7 ± 8.8 vs. 29.6 ± 19.4%, p = 0.029) and higher peak ESR values (36.5 ± 9.7 vs. 25.6 ± 12.8, p < 0.001). Predictors of poor recovery in multivariate logistic regression analysis included age, peak ESR, eGFR, percentage of pulmonary involvement by CT, need for in-hospital oxygen supplementation, SpO2 and mMRC dyspnea score pre-discharge, and history of hypertension. COVID-19 survivors were characterized by decreased physical performance pre-discharge as assessed by the 6 min walk test and did not completely restore their functional status after 1 month of spontaneous recovery, with signs of altered blood oxygenation and dysautonomia contributing to the observed changes. Patients with poor "low base-low increment" trajectory of post-discharge recovery were characterized by younger age but more extensive pulmonary involvement and higher peak ESR values. Poor post-discharge recovery in the study cohort was predictable by the means of machine learning-based classification model that used age, history of hypertension, need for oxygen supplementation, and ESR as inputs.

Machine Learning Based Classification Model for Network Traffic Anomaly Detection

In current days, cloud environments are facing a huge challenge from the attackers in terms of various attacks thrown to the cloud service providers. In both industry and academics, the problem of detection and mitigation of DDoS attacks is now a challenging issue. Detecting Distributed Denial of Service (DDos) threats is mainly a classification problem that can be addressed using data mining, machine learning and deep learning techniques. DDoS attacks can occur in any of the seven-layer OSI model's network. Hence, detecting the DDoS attacks is an important task for cloud service providers to overcome dangerous attacks and loss incurred to stake holders and also the provider.

An Adaptive Deep Learning Neural Network Model to Enhance Machine-Learning-Based Classifiers for Intrusion Detection in Smart Grids

Modern smart grids are built based on top of advanced computing and networking technologies, where condition monitoring relies on secure cyberphysical connectivity. Over the network infrastructure, transported data containing confidential information, must be protected as smart grids are vulnerable and subject to various cyberattacks. Various machine learning based classifiers were proposed for intrusion detection in smart grids. However, each of them has respective advantage and disadvantages. Aiming to improve the performance of existing machine learning based classifiers, this paper proposes an adaptive deep learning algorithm with a data pre-processing module, a neural network pre-training module and a classifier module, which work together classify intrusion data types using their high-dimensional data features. The proposed Adaptive Deep Learning (ADL) algorithm obtains the number of layers and the number of neurons per layer by determining the characteristic dimension of the network traffic. With transfer learning, the proposed ADL algorithm can extract the original data dimensions and obtain new abstract features. By combining deep learning models with traditional machine learning-based classification models, the performance of classification of network traffic data is significantly improved. By using the Network Security Laboratory-Knowledge Discovery in Databases (NSL-KDD) dataset, experimental results show that the proposed ADL algorithm improves the effectiveness of existing intrusion detection methods and reduces the training time, indicating a promising candidate to enhance network security in smart grids.

Early Alzheimer’s Disease Diagnosis Using Wearable Sensors and Multilevel Gait Assessment: A Machine Learning Ensemble Approach

Alzheimer’s disease (AD) is a progressive neurological disorder, and mild cognitive impairment is a stage between cognitive normal and AD. Although timely diagnosis is the key to treatment, the conventional diagnostic methods make periodic diagnosis impossible due to various issues, such as pain and cost. Therefore, we propose a method for early diagnosing by focusing on gait, which is safe and efficient. Seven wearable devices with a built-in inertial measurement unit were used to collect gait data from 145 subjects, and seven gait experiment paradigms, including multilevel subtasks, were developed to clarify the characteristics of gait of each severity. Based on the acquired gait datasets, we proposed a machine learning-based classification model—an elimination method-based ensemble and oversampling model—which is applied to our proposed method. Experimental results show that our proposed model is effective in detecting the early stages of AD and demonstrate the potential of using an auxiliary diagnostic tool.

Predicting students’ academic performance by mining the educational data through machine learning-based classification model

Predicting students’ academic performance by mining the educational data through machine learning-based classification model

Abstract 5362: AI-enabled novel workflow to evaluate T-cell activity in vitro using 3D spheroid models

Abstract T-cell therapies are designed to help our immune system eliminate cancer cells. Those include CAR T-cells (Chimeric Antigen Receptor engineered T-cells), tumor infiltrating lymphocytes (TIL), and other genetically modified T-cells. In recent years, the field of cell therapy has started to expand, including the launch of the first CAR T-cell therapies to treat blood cancer in 2017, which was a critical milestone in this field. Despite its boom, the discovery of novel immunotherapies that specifically enhance T-cell response against cancer cells remains a challenging task, limited by the absence of robust in vitro models to evaluate these immunotherapies throughout their development. The use of CAR T-cells on solid tumors has been lagging due to challenges that include tumor heterogeneity, immunosuppressive microenvironments, and the lack of unique tumor antigens that can be recognized by the CAR-T cells. As such, the ability to screen for CAR T-cells (e.g. with CRISPR) that effectively target and kill tumors is an area of active research. Here we describe a method for assessment of T-cell effect and penetration into the multi-cellular 3D tumor spheroids as a proof-of-concept model for in vitro CAR T assays. Spheroids were formed from Hela cell lines in 96-well round bottom plates. We activated the human peripheral blood mononuclear (PBMC) cells with PMA/ionomycin for 6 hours, then treated the spheroids with stimulated PBMC for 72 hours. Samples were imaged with a confocal imager every 2 hours for a 72 hour period. High-content imaging and analysis allowed us to observe and measure phenotypic changes in cancer spheroids and the process of T-cell penetration over a period of 72 hours. To optimize the workflow, we developed an image analysis approach that uses a deep learning-based segmentation model and a machine learning-based classification model to quantify the T-cell induced phenotypic changes in the spheroids using brightfield images. Our results show clear distinct phenotypic changes among different treatment groups and the feasibility of using AI-based analysis workflows to accurately predict the efficacy of T-cells in an in vitro assay. Moreover, the information associated with the penetration of T-cell into the 3D spheroids was explored by calculating the coordinates of the penetrated T-cells and T-cell movements from the nearest edge of the spheroids where the T-cell penetration speed was estimated. Our results show that the stimulated T-cells have much greater penetration speed compared to non-stimulated T-cells. Image analysis also allowed us to measure and quantitate the deterioration of cancer spheroids and changes in their morphology. Overall, our results show that the 3D spheroid models and the high-content analysis workflow may potentially be used as a metric to evaluate the efficacy of cell therapy in vitro. Citation Format: Zhisong Tong, Oksana Sirenko, Misha Bashkurov, Angeline Lim. AI-enabled novel workflow to evaluate T-cell activity in vitro using 3D spheroid models. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2023; Part 1 (Regular and Invited Abstracts); 2023 Apr 14-19; Orlando, FL. Philadelphia (PA): AACR; Cancer Res 2023;83(7_Suppl):Abstract nr 5362.

Applying multisequence MRI radiomics of the primary tumor and lymph node to predict HPV-related p16 status in patients with oropharyngeal squamous cell carcinoma.

Human papillomavirus (HPV)-related oropharyngeal squamous cell carcinoma (OPSCC) is currently rising in incidence; however, a noninvasive approach to predicting HPV status that strongly correlates with p16 expression is lacking. This study aimed to develop a radiomics model based on multisequence magnetic resonance imaging (MRI) of primary tumor (PT) and lymph node (LN)-fused imaging features for the prediction of p16 status in OPSCC. In this retrospective study, 141 patients (comprising 116 patients in the training cohort and 25 patients in the testing cohort) with histopathologically confirmed (OPSCC) were enrolled consecutively from Fudan University Shanghai Cancer Center and Shanghai Ninth People's Hospital between January 2011 and December 2020. HPV status was determined by p16 immunohistochemistry analysis. A total of 2092 radiomics features were initially computed and extracted from the 3D-segmented PT and largest LN based on contrast-enhanced T1-weighted imaging (CE-T1WI) and T2-weighted imaging (T2WI). A support vector machine classifier was employed to build the machine learning-based classification models dependent on p16 status. The models were validated in the testing cohort. The area under the receiver operating characteristic curve (AUC) was computed to assess the performance of each model. This diagnostic study was not registered on the clinical trial platform. In the testing cohort, fusion models yielded better performance (AUC) compared with models based on a sole PT/LN [CE-T1WI: 0.80 (95% CI: 0.55-0.94) vs. 0.71 (95% CI: 0.47-0.88)/0.73 (95% CI: 0.48-0.90); T2WI: 0.74 (95% CI: 0.51-0.95) vs. 0.64 (95% CI: 0.38-0.85)/0.71 (95% CI: 0.48-0.88)]. Models based on multisequence imaging outperformed single CE-T1WI/T2WI models [PT: 0.74 (95% CI: 0.46-0.91) vs. 0.71 (95% CI: 0.47-0.88)/0.64 (95% CI: 0.38-0.85); LN: 0.78 (95% CI: 0.55-0.75) vs. 0.73 (95% CI: 0.48-0.90)/0.71 (95% CI: 0.48-0.88)]. Finally, the PT-LN fusion model based on multisequencing yielded the best classification performance with the highest AUC value of 0.91 (95% CI: 0.72-0.98) for the prediction of p16 expression. The differences between the performance of the final model and the other 8 models were significant (all P values <0.05). The results demonstrated that (I) the PT-LN fusion radiomics models improved the classification performance of the sole use of PT or LN for the prediction of p16 status, (II) the radiomics models based on multisequences outperformed the single-sequence models in the prediction of p16 status, and (III) the PT-LN fusion model based on multisequence MRI radiomics features could serve as a noninvasive method for acquiring the molecular information of patients with OPSCC, potentially assisting oncologists with their clinical decision-making.