Multiparametric MRI Radiomics With Machine Learning for Differentiating HER2-Zero, -Low, and -Positive Breast Cancer: Model Development, Testing, and Interpretability Analysis

American Journal of Roentgenology, Journal Year: 2024, Volume and Issue: 224(1)

Published: Oct. 16, 2024

BACKGROUND. MRI radiomics has been explored for three-tiered classification of HER2 expression levels (i.e., HER2-zero, HER2-low, or HER2-positive) in patients with breast cancer, although an understanding how such models reach their predictions is lacking. OBJECTIVE. The purpose this study was to develop and test multiparametric machine learning differentiating as well explain the contributions model features through local global interpretations use Shapley additive explanation (SHAP) analysis. METHODS. This retrospective included 737 (mean age, 54.1 ± 10.6 [SD] years) cancer from two centers (center 1 [n = 578] center 2 159]), all whom underwent had determined after excisional biopsy. Analysis entailed tasks: HER2-negative HER2-zero HER2-low) tumors HER2-positive (task 1) HER2-low 2). For each task, were randomly assigned a 7:3 ratio training set 1: n 405; task 2: 284) internal 173; 122); formed external 159; 105). Radiomic extracted early phase dynamic contrast-enhanced (DCE) imaging, T2-weighted DWI. support vector (SVM) used feature selection, score (radscore) computed using weights SVM correlation coefficients, conventional combined constructed, performances evaluated. SHAP analysis provide outputs. RESULTS. In set, 1, AUCs model, radscore, 0.624, 0.757, 0.762, respectively; 2, AUC radscore 0.754, no could be constructed. identified DCE imaging having strongest influence both tasks; also prominent role 2. CONCLUSION. findings indicate suboptimal performance noninvasive characterization expression. CLINICAL IMPACT. provides example interpretation better understand imaging-based models.

Language: Английский

---

ChatGPT as an effective tool for quality evaluation of radiomics research

European Radiology, Journal Year: 2024, Volume and Issue: unknown

Published: Oct. 15, 2024

Language: Английский

---

Classifying the molecular subtype of breast cancer using vision transformer and convolutional neural network features

Breast Cancer Research and Treatment, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 22, 2025

Language: Английский

---

Intratumoral and peritumoral ultrasound-based radiomics for preoperative prediction of HER2-low breast cancer: a multicenter retrospective study

Insights into Imaging, Journal Year: 2025, Volume and Issue: 16(1)

Published: March 7, 2025

Recent advances in human epidermal growth factor receptor 2 (HER2)-targeted therapies have opened up new therapeutic options for HER2-low cancers. This study aimed to establish an ultrasound-based radiomics model identify three different HER2 states noninvasively. Between May 2018 and December 2023, a total of 1257 invasive breast cancer patients were enrolled from hospitals. The status was divided into classes: positive, low, zero. Four peritumoral regions interest (ROI) auto-generated by dilating the manually segmented intratumoral ROI thicknesses 5 mm, 10 15 20 mm. After image preprocessing, 4720 features extracted each every patient. least absolute shrinkage selection operator LightBoost algorithm utilized construct single- multi-region signatures (RS). A clinical-radiomics combined developed integrating discriminative clinical-sonographic factors with optimal RS. data stitching strategy used build patient-level models. Shapley additive explanations (SHAP) approach explain contribution internal prediction. RS constructed 12 tumor 9 peritumoral-15mm features. Age, size, seven qualitative ultrasound retained In training, validation, test cohorts, showed best discrimination ability macro-AUCs 0.988 (95% CI: 0.983-0.992), 0.915 0.851-0.965), 0.862 0.820-0.899), respectively. built robust interpretable evaluate classes based on images. Ultrasound-based method can noninvasively HER2, which may guide treatment decisions implementation personalized HER2-targeted patients. Determination affect cancer. discriminate statuses. Our assist providing recommendations novel therapies.

Language: Английский

---

Microstructural diffusion MRI for differentiation of breast tumors and prediction of prognostic factors in breast cancer

Frontiers in Oncology, Journal Year: 2025, Volume and Issue: 15

Published: March 5, 2025

This study aims to investigate the feasibility of cellular microstructural mapping by diffusion MRI (IMPULSED, imaging parameters using limited spectrally edited diffusion) breast tumors, and further evaluate whether MRI-derived features is associated with prognostic factors in cancer. prospective collected 232 patients suspected tumors from March August 2023. The IMPULSED scan included acquisitions both pulsed (PGSE) oscillating (OGSE) gradient spin echo frequencies up 33 Hz. OGSE PGSE data were fitted IMPUSLED method a two-compartment model estimate mean cell diameter (d mean), intracellular fraction (fin ), extracellular diffusivity (D ex), cellularity index (f in/d) within tumor lesions. apparent coefficients (ADCs) calculated conventional weighted imaging, PGSE, (17 Hz Hz) sequences (ADCDWI, ADCPGSE, ADC17Hz, ADC33Hz). independent samples test was used compare d mean, fin , Dex index, ADC values between benign malignant cancer subgroups different risk factors. receiver operating characteristic (ROC) curve access diagnostic performance. 213 finally divided into (n=130) (n=83) groups according histopathological results. (15.74 ± 2.68 vs. 14.28 4.65 μm, p<0.001), f (0.346 0.125 0.279 0.212, p<0.001) (21.19 39.54 19.38 14.87 ×10-3 um-1, p<0.005) lesions significantly higher than those lesions, D ex (2.119 0.395 2.378 0.332 um2/ms, ADCDWI (0.877 0.148 1.453 0.356 lower For differentiation showed highest AUC 0.951 sensitivity 80.49% specificity 98.28%. combination in, ex, for 0.787 (sensitivity = 70.73%, 77.86%), IMPULSED-derived ADCs improve 0.897 81.93%, 81.54%). HER-2(+) HER-2(-) (0.313 0.100 0.371 0.137, p=0.015), ADCDWI, ADC17Hz ADC33Hz (ADCDWI: 0.929 0.115 0.855 0.197 p=0.023; ADC17Hz: 1.373 0.306 1.242 0.301 um2/s, p =0.025; ADC33Hz: 2.042 0.545 1.811 0.392 0.008). (0.377 0.136 0.300 0.917, p=0.001) (27.22 12.02 21.66 7.76 p=0.007) PR(+) PR(-) tumor. tumors(1.227 0.299 1.404 0.294 =0.002).The ER(+) ER(-) (ADC17Hz: 1.258 0.313 1.400 0.273 0.029; ex: 2.070 0.405 2.281 0.331 p=0.011). ER(-), AUCs 0.643 76.67%, 47.06%) 0.646 80.0%, 45.98%), 0.663 =93.33%, 36.78%). PR(-), 0.666 68.18%, 61.97%), 0.697 77.27%, 60.27%) 0.661 61.64%), respectively, their 0.729 =72.73%, 65.75%). HER-2(-), ADC33Hz, 0.625 59.42%, 63.04%), 0.632 43.66%, 84.78%), 0.664 47.95%, 82.67%) 0.650 77.46%, 56.52%), 0.693 69.57%, 64.79%) HER-2(-). demonstrates promise characterizing which may be helpful evaluation

Language: Английский

---

Synthetic MRI, dynamic contrast-enhanced MRI combined with diffusion-weighted imaging for identifying molecular subtypes of breast cancer using machine learning models

Research Square (Research Square), Journal Year: 2024, Volume and Issue: unknown

Published: Dec. 26, 2024

Objective: To determine whether quantitative parameters from synthetic magnetic resonance imaging (SyMRI), dynamic contrast-enhanced MRI (DCE-MRI), and diffusion-weighted (DWI) can effectively differentiate between molecular subtypes of breast cancer using various machine learning models. Materials Methods: This retrospective study included 401 patients with suspicious lesions who underwent examinations, including SyMRI, DCE-MRI, DWI, September 2020 to 2024. Quantitative obtained SyMRI T1-Pre, T2-Pre, proton density (PD-Pre) values before contrast injection, as well T1-Gd, T2-Gd, PD-Gd after injection. Additionally, difference (Delta-T1, Delta-T2, Delta-PD) enhancement ratios (T1-Ratio, T2-Ratio, PD-Ratio) were calculated. Two radiologists retrospectively evaluated the morphological kinetic characteristics on apparent diffusion coefficient (ADC) assess tumors DWI. Logistic regression ANOVA applied identify significant parameter differences among four subtypes. Based these selected by logistic regression, five models developed: Regression (LR), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Random Forest (RF), Decision Tree (DT). We plotted Receiver Operating Characteristic (ROC) curves calculated area under curve (AUC) primary metric performance best model. utilized SHAP library in Python generate feature importance for our model's predictions. Results: A total 292 (median age, 53 years; age range, 27–80 years) met inclusion criteria. Among these, 204 52 27–78 assigned training cohort, while 88 testing cohort. Eleven identified across subtypes(p＜0.05). These two clinical pathological factors: menopause(p＜0.001); parameters: PD-Gd, T1-Ratio, PD-Ratio(p＜0.05); three DCE-MRI burr sign, time–intensity (TIC), Breast Imaging Reporting Date System(BI-RADS) grading(p＜0.001); one DWI parameter: ADC-Tumor(p＜0.001). The SVM model demonstrated highest overall based comprehensive evaluation multiple metrics set, achieving superior diagnostic AUC, accuracy, specificity, sensitivity 0.972, 82.5%, 94.76%, 82.14%, respectively. achieved AUC 0.979 luminal A, 0.925 B, 0.971 HER2-enriched, 0.982 triple-negative (TN) set; 0.973 0.873 0.956 0.955 TN set. Shapley Additive Explanations (SHAP) tool features contributing model, PD-Ratio, sign showing contributions, mean absolute 0.418, 0.340, 0.264, Conclusion: derived mappings, may provide a non-invasive approach differentiating

Language: Английский

---

Clinical-radiomics nomogram based on the fat-suppressed T2 sequence for differentiating luminal and non-luminal breast cancer

Frontiers in Oncology, Journal Year: 2024, Volume and Issue: 14

Published: Oct. 25, 2024

To establish and validate a new clinical-radiomics nomogram based on the fat-suppressed T2 sequence for differentiating luminal non-luminal breast cancer.

Language: Английский

---