Radiomics and Deep Features: Robust Classification of Brain Hemorrhages and Reproducibility Analysis Using a 3D Autoencoder Neural Network

Bioengineering, Journal Year: 2024, Volume and Issue: 11(7), P. 643 - 643

Published: June 24, 2024

This study evaluates the reproducibility of machine learning models that integrate radiomics and deep features (features extracted from a 3D autoencoder neural network) to classify various brain hemorrhages effectively. Using dataset 720 patients, we 215 (RFs) 15,680 (DFs) CT images. With rigorous screening based on Intraclass Correlation Coefficient thresholds (>0.75), identified 135 RFs 1054 DFs for analysis. Feature selection techniques such as Boruta, Recursive Elimination (RFE), XGBoost, ExtraTreesClassifier were utilized alongside 11 classifiers, including AdaBoost, CatBoost, Decision Trees, LightGBM, Logistic Regression, Naive Bayes, Neural Networks, Random Forest, Support Vector Machines (SVM), k-Nearest Neighbors (k-NN). Evaluation metrics included Area Under Curve (AUC), Accuracy (ACC), Sensitivity (SEN), F1-score. The model evaluation involved hyperparameter optimization, 70:30 train–test split, bootstrapping, further validated with Wilcoxon signed-rank test q-values. Notably, showed higher accuracy. In case RFs, Boruta + SVM combination emerged optimal AUC, ACC, SEN, while XGBoost Forest excelled in Specifically, achieved F1-scores 0.89, 0.85, 0.82, 0.80, respectively. Among DFs, Bayes demonstrated remarkable performance, attaining an AUC 0.96, ACC 0.93, SEN 0.92, F1-score 0.92. Distinguished RF category Regression ExtraTreesClassifier, CatBoost each yielding significant q-values 42. realm, k-NN exhibited robustness, 43, 41 q-values, investigation underscores potential synergizing serve valuable tools, thereby enhancing interpretation head scans patients hemorrhages.

Language: Английский

---

Artificial intelligence performance in detecting lymphoma from medical imaging: a systematic review and meta-analysis

BMC Medical Informatics and Decision Making, Journal Year: 2024, Volume and Issue: 24(1)

Published: Jan. 8, 2024

Abstract Background Accurate diagnosis and early treatment are essential in the fight against lymphatic cancer. The application of artificial intelligence (AI) field medical imaging shows great potential, but diagnostic accuracy lymphoma is unclear. This study was done to systematically review meta-analyse researches concerning performance AI detecting using for first time. Methods Searches were conducted Medline, Embase, IEEE Cochrane up December 2023. Data extraction assessment included quality independently by two investigators. Studies that reported an model/s detection systemic review. We extracted binary data obtain outcomes interest: sensitivity (SE), specificity (SP), Area Under Curve (AUC). registered with PROSPERO, CRD42022383386. Results Thirty studies systematic review, sixteen which meta-analyzed a pooled 87% (95%CI 83–91%), 94% (92–96%), AUC 97% (95–98%). Satisfactory observed subgroup analyses based on algorithms types (machine learning versus deep learning, whether transfer applied), sample size (≤ 200 or > 200), clinicians models geographical distribution institutions (Asia non-Asia). Conclusions Even if possible overestimation further better standards needed, we suggest may be useful diagnosis.

Language: Английский

---

The value of machine learning approaches in the diagnosis of early gastric cancer: a systematic review and meta-analysis

World Journal of Surgical Oncology, Journal Year: 2024, Volume and Issue: 22(1)

Published: Feb. 1, 2024

Abstract Background The application of machine learning (ML) for identifying early gastric cancer (EGC) has drawn increasing attention. However, there lacks evidence-based support its specific diagnostic performance. Hence, this systematic review and meta-analysis was implemented to assess the performance image-based ML in EGC diagnosis. Methods We performed a comprehensive electronic search PubMed, Embase, Cochrane Library, Web Science up September 25, 2022. QUADAS-2 selected judge risk bias included articles. did using bivariant mixed-effect model. Sensitivity analysis heterogeneity test were performed. Results Twenty-one articles enrolled. sensitivity (SEN), specificity (SPE), SROC ML-based models 0.91 (95% CI: 0.87–0.94), 0.85 0.81–0.89), 0.94 0.39–1.00) training set 0.90 0.86–0.93), 0.86–0.92), 0.96 0.19–1.00) validation set. SEN, SPE, diagnosis by non-specialist clinicians 0.64 0.56–0.71), 0.84 0.77–0.89), 0.80 0.29–0.97), those specialist 0.74–0.85), 0.88 0.85–0.91), 0.37–0.99). With assistance models, SEN physicians significantly improved (0.76 vs 0.64). Conclusion have greater identification EGC. accuracy can be level specialists with models. results suggest that better assist less experienced diagnosing under endoscopy broad clinical value.

Language: Английский

---

Comparison of artificial intelligence and logistic regression models for mortality prediction in acute respiratory distress syndrome: a systematic review and meta-analysis

Intensive Care Medicine Experimental, Journal Year: 2025, Volume and Issue: 13(1)

Published: Feb. 21, 2025

Abstract Background The application of artificial intelligence (AI) in predicting the mortality acute respiratory distress syndrome (ARDS) has garnered significant attention. However, there is still a lack evidence-based support for its specific diagnostic performance. Thus, this systematic review and meta-analysis was conducted to evaluate effectiveness AI algorithms ARDS mortality. Method We comprehensive electronic search across Web Science, Embase, PubMed, Scopus , EBSCO databases up April 28, 2024. QUADAS-2 tool used assess risk bias included articles. A bivariate mixed-effects model applied meta-analysis. Sensitivity analysis, meta-regression tests heterogeneity were also performed. Results Eight studies analysis. sensitivity, specificity, summarized receiver operating characteristic (SROC) AI-based validation set 0.89 (95% CI 0.79–0.95), 0.72 0.65–0.78), 0.84 0.80–0.87), respectively. For logistic regression (LR) model, SROC 0.78 0.74–0.82), 0.68 0.60–0.76), 0.81 0.77–0.84). demonstrated superior predictive accuracy compared LR model. Notably, performed better patients with moderate severe (SAUC: [95% 0.80–0.87] vs. 0.77–0.84]). Conclusion showed performance strong potential clinical application. Additionally, we found that ARDS, highly heterogeneous condition, influenced by severity disease.

Language: Английский

---

A PET/CT radiomics model for predicting distant metastasis in early-stage non–small cell lung cancer patients treated with stereotactic body radiotherapy: a multicentric study

Radiation Oncology, Journal Year: 2024, Volume and Issue: 19(1)

Published: Jan. 22, 2024

Abstract Objectives Stereotactic body radiotherapy ( SBRT) is a treatment option for patients with early-stage non-small cell lung cancer (NSCLC) who are unfit surgery. Some may experience distant metastasis. This study aimed to develop and validate radiomics model predicting metastasis in NSCLC treated SBRT. Methods Patients at five institutions were enrolled this study. Radiomics features extracted based on the PET/CT images. After feature selection training set (from Tianjin), CT-based PET-based signatures built. Models CT PET built validated using external datasets Zhejiang, Zhengzhou, Shandong, Shanghai). An integrated that included radiomic was developed. The performance of proposed evaluated terms its discrimination, calibration, clinical utility. Multivariate logistic regression used calculate probability metastases. cutoff value obtained receiver operator characteristic curve (ROC), divided into high- low-risk groups. Kaplan-Meier analysis evaluate metastasis-free survival (DMFS) different risk Results In total, 228 enrolled. median follow-up time 31.4 (2.0-111.4) months. had an area under (AUC) 0.819 n = 139) 0.786 dataset 89). AUC 0.763 0.804 dataset. combining 0.835 combined showed moderate calibration positive net benefit. When greater than 0.19, patient considered be high risk. DMFS significantly stratified P < 0.001). Conclusions can predict SBRT provide reference decision-making. Plain language summary study, established by moderate-quantity cohort successfully independent cohorts. Physicians could use easy-to-use assess after Identifying subgroups factors useful guiding personalized approaches.

Language: Английский

---

Intra- and peritumoral radiomics features based on multicenter automatic breast volume scanner for noninvasive and preoperative prediction of HER2 status in breast cancer: a model ensemble research

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

Published: Feb. 29, 2024

Abstract The aim to investigate the predictive efficacy of automatic breast volume scanner (ABVS), clinical and serological features alone or in combination at model level for predicting HER2 status. weighted method was developed identify status compared with single data source feature method. 271 patients invasive cancer were included retrospective study, which 174 our center randomized into training validation sets, 97 external as test set. Radiomics extracted from ABVS-based tumor, peritumoral 3 mm region, 5 region used construct four types optimal models, Tumor, R3mm, R5mm, Clinical model, respectively. Then, methods performed optimize models. proposed models achieved better performance both set For set, highest area under curve (AUC) 0.803 (95% confidence interval [CI] 0.660–947), 0.739 (CI 0.556,0.921), 0.826 CI 0.689,0.962), respectively; sensitivity specificity 100%, 62.5%; 81.8%, 66.7%; 90.9%,75.0%; attained best AUC 0.695 0.583, 0.807), 0.668 0.555,0.782), 0.700 0.590,0.811), 86.1%, 41.9%; 61.1%, 71.0%; a model. optimized composed intratumoral radiomics may be potential biomarkers noninvasive preoperative prediction cancer.

Language: Английский

---

Application of machine learning-based multi-sequence MRI radiomics in diagnosing anterior cruciate ligament tears

Journal of Orthopaedic Surgery and Research, Journal Year: 2024, Volume and Issue: 19(1)

Published: Jan. 31, 2024

To compare the diagnostic power among various machine learning algorithms utilizing multi-sequence magnetic resonance imaging (MRI) radiomics in detecting anterior cruciate ligament (ACL) tears. Additionally, this research aimed to create and validate optimal model.

Language: Английский

---

Prediction of the Ki-67 expression level in head and neck squamous cell carcinoma with machine learning-based multiparametric MRI radiomics: a multicenter study

BMC Cancer, Journal Year: 2024, Volume and Issue: 24(1)

Published: April 5, 2024

Abstract Background This study aimed to develop and validate a machine learning (ML)-based fusion model preoperatively predict Ki-67 expression levels in patients with head neck squamous cell carcinoma (HNSCC) using multiparametric magnetic resonance imaging (MRI). Methods A total of 351 pathologically proven HNSCC from two medical centers were retrospectively enrolled the divided into training ( n = 196), internal validation 84), external 71) cohorts. Radiomics features extracted T2-weighted images contrast-enhanced T1-weighted screened. Seven ML classifiers, including k-nearest neighbors (KNN), support vector (SVM), logistic regression (LR), random forest (RF), linear discriminant analysis (LDA), naive Bayes (NB), eXtreme Gradient Boosting (XGBoost) trained. The best classifier was used calculate radiomics (Rad)-scores combine clinical factors construct model. Performance evaluated based on calibration, discrimination, reclassification, utility. Results Thirteen combining MRI finally selected. SVM showed performance, highest average area under curve (AUC) 0.851 incorporating SVM-based Rad-scores T stage MR-reported lymph node status achieved encouraging predictive performance (AUC 0.916), 0.903), 0.885) Furthermore, better benefit higher classification accuracy than Conclusions ML-based exhibited promise for predicting patients, which might be helpful prognosis evaluation decision-making.

Language: Английский

---

Clinical feasibility of deep learning-based synthetic CT images from T2-weighted MR images for cervical cancer patients compared to MRCAT

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

Published: April 12, 2024

Abstract This work aims to investigate the clinical feasibility of deep learning-based synthetic CT images for cervix cancer, comparing them MR calculating attenuation (MRCAT). Patient cohort with 50 pairs T2-weighted and from cervical cancer patients was split into 40 training 10 testing phases. We conducted deformable image registration Nyul intensity normalization maximize similarity between as a preprocessing step. The processed were plugged learning model, generative adversarial network. To prove feasibility, we assessed accuracy in using structural (SSIM) mean-absolute-error (MAE) dosimetry gamma passing rate (GPR). Dose calculation performed on true commercial Monte Carlo algorithm. Synthetic generated by outperformed MRCAT 1.5% SSIM, 18.5 HU MAE. In dosimetry, DL-based achieved 98.71% 96.39% GPR at 1% 1 mm criterion 10% 60% cut-off values prescription dose, which 0.9% 5.1% greater GPRs over images.

Language: Английский

---

Deep learning for risk stratification of thymoma pathological subtypes based on preoperative CT images

BMC Cancer, Journal Year: 2024, Volume and Issue: 24(1)

Published: May 28, 2024

Abstract Objectives This study aims to develop an innovative, deep model for thymoma risk stratification using preoperative CT images. Current algorithms predominantly focus on radiomic features or 2D and require manual tumor segmentation by radiologists, limiting their practical applicability. Methods The was trained tested a dataset comprising images from 147 patients (82 female; mean age, 54 years ± 10) who underwent surgical resection received subsequent pathological confirmation. eligible participants were divided into training cohort (117 patients) testing (30 based the scan time. consists of two stages: 3D stratification. (2D) constructed comparative analysis. Model performance evaluated through dice coefficient, area under curve (AUC), accuracy. Results In both cohorts, demonstrated better in differentiating risk, boasting AUCs 0.998 0.893 respectively. compared (AUCs 0.773 0.769) 0.981 0.760). Notably, capable simultaneously identifying lesions, segmenting region interest (ROI), arterial phase Its diagnostic prowess outperformed that baseline model. Conclusions has potential serve as innovative decision-making tool, assisting clinical prognosis evaluation discernment suitable treatments different subtypes. Key Points • incorporated model, features, effectively predicted risk. improved 16.1pt 17.5pt

Language: Английский

---