Predicting lung cancer survival based on clinical data using machine learning: A review

Computers in Biology and Medicine, Journal Year: 2023, Volume and Issue: 165, P. 107338 - 107338

Published: Aug. 9, 2023

Machine learning has gained popularity in predicting survival time the medical field. This review examines studies utilizing machine and data-mining techniques to predict lung cancer using clinical data. A systematic literature searched MEDLINE, Scopus, Google Scholar databases, following reporting guidelines COVIDENCE system. Studies published from 2000 2023 employing for prediction were included. Risk of bias assessment used model risk tool. Thirty reviewed, with 13 (43.3%) surveillance, epidemiology, end results database. Missing data handling was addressed 12 (40%) studies, primarily through transformation conversion. Feature selection algorithms 19 (63.3%) age, sex, N stage being most chosen features. Random forest predominant model, 17 (56.6%) studies. While number is limited, use models based on grown since 2012. Consideration diverse patient cohorts pre-processing are crucial. Notably, did not account missing data, normalization, scaling, or standardized potentially introducing bias. Therefore, a comprehensive study needed, addressing these challenges.

Language: Английский

---

Publicly available datasets of breast histopathology H&E whole-slide images: A scoping review

Journal of Pathology Informatics, Journal Year: 2024, Volume and Issue: 15, P. 100363 - 100363

Published: Feb. 1, 2024

Advancements in digital pathology and computing resources have made a significant impact the field of computational for breast cancer diagnosis treatment. However, access to high-quality labeled histopathological images is big challenge that limits development accurate robust deep learning models. In this scoping review, we identified publicly available datasets H&E-stained whole-slide (WSIs) can be used develop algorithms. We systematically searched 9 scientific literature databases research data repositories found 17 containing 10 385 H&E WSIs cancer. Moreover, reported image metadata characteristics each dataset assist researchers selecting proper specific tasks pathology. addition, compiled 2 lists patches private as supplementary researchers. Notably, only 28% included articles utilized multiple datasets, 14% an external validation set, suggesting performance other developed models may susceptible overestimation. The TCGA-BRCA was 52% selected studies. This has considerable selection bias robustness generalizability trained There also lack consistent reporting WSI issue developing models, indicating necessity establishing explicit guidelines documenting metadata.

Language: Английский

---

Multimodal artificial intelligence-based pathogenomics improves survival prediction in oral squamous cell carcinoma

Scientific Reports, Journal Year: 2024, Volume and Issue: 14(1)

Published: March 7, 2024

Abstract In this study, we aimed to develop a novel prognostic algorithm for oral squamous cell carcinoma (OSCC) using combination of pathogenomics and AI-based techniques. We collected comprehensive clinical, genomic, pathology data from cohort OSCC patients in the TCGA dataset used machine learning deep algorithms identify relevant features that are predictive survival outcomes. Our analyses included 406 patients. Initial involved gene expression analyses, principal component enrichment feature importance analyses. These insights were foundational subsequent model development. Furthermore, applied five learning/deep (Random Survival Forest, Gradient Boosting Analysis, Cox PH, Fast SVM, DeepSurv) prediction. initial revealed variations biological pathways, laying groundwork robust selection building. The results showed multimodal outperformed unimodal models across all methods, with c-index values 0.722 RSF, 0.633 GBSA, 0.625 FastSVM, CoxPH, 0.515 DeepSurv. When considering only important features, continued outperform models, 0.834 0.747 0.718 0.742 0.635 demonstrate potential techniques improving accuracy prediction OSCC, which may ultimately aid development personalized treatment strategies devastating disease.

Language: Английский

---

A Novel Approach for Predicting the Survival of Colorectal Cancer Patients Using Machine Learning Techniques and Advanced Parameter Optimization Methods

Cancers, Journal Year: 2024, Volume and Issue: 16(18), P. 3205 - 3205

Published: Sept. 20, 2024

Background: Colorectal cancer is one of the most prevalent forms and associated with a high mortality rate. Additionally, an increasing number adults under 50 are being diagnosed disease. This underscores importance leveraging modern technologies, such as artificial intelligence, for early diagnosis treatment support. Methods: Eight classifiers were utilized in this research: Random Forest, XGBoost, CatBoost, LightGBM, Gradient Boosting, Extra Trees, k-nearest neighbor algorithm (KNN), decision trees. These algorithms optimized using frameworks Optuna, RayTune, HyperOpt. study was conducted on public dataset from Brazil, containing information tens thousands patients. Results: The models developed demonstrated classification accuracy predicting one-, three-, five-year survival, well overall cancer-specific mortality. Forest delivered best performance, achieving approximately 80% across all evaluated tasks. Conclusions: research enabled development effective that can be applied clinical practice.

Language: Английский

---

Transformer enabled multi-modal medical diagnosis for tuberculosis classification

Journal Of Big Data, Journal Year: 2025, Volume and Issue: 12(1)

Published: Jan. 14, 2025

Abstract Recently, multimodal data analysis in medical domain has started receiving a great attention. Researchers from both computer science, and medicine are trying to develop models handle data. However, most of the published work have targeted homogeneous The collection preparation heterogeneous is complex time-consuming task. Further, development such another challenge. This study presents cross modal transformer-based fusion approach for clinical using images proposed leverages image embedding layer convert into visual tokens, text tokens. cross-modal transformer module employed learn holistic representation imaging modalities. was tested multi-modal lung disease tuberculosis set. results compared with recent approaches field analysis. comparison shows that outperformed other considered study. Another advantage this it faster analyze existing methods used study, which very important if we do not powerful machines computation.

Language: Английский

---

Artificial intelligence and machine learning in cell-free-DNA-based diagnostics

Genome Research, Journal Year: 2025, Volume and Issue: 35(1), P. 1 - 19

Published: Jan. 1, 2025

The discovery of circulating fetal and tumor cell-free DNA (cfDNA) molecules in plasma has opened up tremendous opportunities noninvasive diagnostics such as the detection chromosomal aneuploidies cancers posttransplantation monitoring. advent high-throughput sequencing technologies makes it possible to scrutinize characteristics cfDNA molecules, opening fields genetics, epigenetics, transcriptomics, fragmentomics, providing a plethora biomarkers. Machine learning (ML) and/or artificial intelligence (AI) that are known for their ability integrate high-dimensional features have recently been applied field liquid biopsy. In this review, we highlight various AI ML approaches cfDNA-based diagnostics. We first introduce biology basic concepts technologies. then discuss selected examples ML- or AI-based applications prenatal testing cancer These include deduction fraction, tissue mapping, localization. Finally, offer perspectives on future direction using leverage fragmentation patterns terms methylomic transcriptional investigations.

Language: Английский

---

Comparing the Effectiveness of Artificial Intelligence Models in Predicting Ovarian Cancer Survival: A Systematic Review

Cancer Reports, Journal Year: 2025, Volume and Issue: 8(3)

Published: March 1, 2025

ABSTRACT Background This systematic review investigates the use of machine learning (ML) algorithms in predicting survival outcomes for ovarian cancer (OC) patients. Key prognostic endpoints, including overall (OS), recurrence‐free (RFS), progression‐free (PFS), and treatment response prediction (TRP), are examined to evaluate effectiveness these identify significant features that influence predictive accuracy. Recent Findings A thorough search four major databases—PubMed, Scopus, Web Science, Cochrane—resulted 2400 articles published within last decade, with 32 studies meeting inclusion criteria. Notably, most publications emerged after 2021. Commonly used included random forest, support vector machines, logistic regression, XGBoost, various deep models. Evaluation metrics such as area under curve (AUC) (18 studies), concordance index (C‐index) (11 accuracy studies) were frequently employed. Age at diagnosis, tumor stage, CA‐125 levels, treatment‐related factors consistently highlighted predictors, emphasizing their relevance OC prognosis. Conclusion ML models demonstrate considerable potential outcomes; however, challenges persist regarding model interpretability. Incorporating diverse data types—such clinical, imaging, molecular datasets—holds promise enhancing capabilities. Future advancements will depend on integrating heterogeneous sources multimodal approaches, which crucial improving precision OC.

Language: Английский

---

Combining Molecular, Imaging, and Clinical Data Analysis for Predicting Cancer Prognosis

Cancers, Journal Year: 2022, Volume and Issue: 14(13), P. 3215 - 3215

Published: June 30, 2022

Cancer is one of the most detrimental diseases globally. Accordingly, prognosis prediction cancer patients has become a field interest. In this review, we have gathered 43 state-of-the-art scientific papers published in last 6 years that built predictive models using multimodal data. We defined multimodality data as four main types: clinical, anatomopathological, molecular, and medical imaging; expanded on information each modality provides. The studies were divided into three categories based modelling approach taken, their characteristics further discussed together with current issues future trends. Research area evolved from survival analysis through statistical mainly clinical anatomopathological to multi-faceted data-driven by integration complex, multimodal, high-dimensional containing multi-omics imaging applying Machine Learning and, more recently, Deep techniques. This review concludes are capable better stratifying patients, which can improve management contribute implementation personalised medicine well provide new valuable knowledge biology its progression.

Language: Английский

---

Hybrid Graph Convolutional Network With Online Masked Autoencoder for Robust Multimodal Cancer Survival Prediction

IEEE Transactions on Medical Imaging, Journal Year: 2023, Volume and Issue: 42(8), P. 2462 - 2473

Published: March 6, 2023

Cancer survival prediction requires exploiting related multimodal information (e.g., pathological, clinical and genomic features, etc.) it is even more challenging in practices due to the incompleteness of patient's data. Furthermore, existing methods lack sufficient intra- inter-modal interactions, suffer from significant performance degradation caused by missing modalities. This manuscript proposes a novel hybrid graph convolutional network, entitled HGCN, which equipped with an online masked autoencoder paradigm for robust cancer prediction. Particularly, we pioneer modeling data into flexible interpretable graphs modality-specific preprocessing. HGCN integrates advantages networks (GCNs) hypergraph network (HCN) through node message passing hyperedge mixing mechanism facilitate intra-modal interactions between graphs. With potential create reliable predictions risk dramatically increased compared prior methods. Most importantly, compensate patient modalities scenarios, incorporated can effectively capture intrinsic dependence seamlessly generate hyperedges model inference. Extensive experiments analysis on six cohorts TCGA show that our method significantly outperforms state-of-the-arts both complete modal settings. Our codes are made available at https://github.com/lin-lcx/HGCN.

Language: Английский

---

Automatic Segmentation with Deep Learning in Radiotherapy

Cancers, Journal Year: 2023, Volume and Issue: 15(17), P. 4389 - 4389

Published: Sept. 1, 2023

This review provides a formal overview of current automatic segmentation studies that use deep learning in radiotherapy. It covers 807 published papers and includes multiple cancer sites, image types (CT/MRI/PET), methods. We collect key statistics about the to uncover commonalities, trends, methods, identify areas where more research might be needed. Moreover, we analyzed corpus by posing explicit questions aimed at providing high-quality actionable insights, including: “What should researchers think when starting study?”, “How can practices medical improved?”, is missing from corpus?”, more. allowed us provide practical guidelines on how conduct good study today’s competitive environment will useful for future within field, regardless specific radiotherapeutic subfield. To aid our analysis, used large language model ChatGPT condense information.

Language: Английский

---