Table 4.
Summary of representative studies on 18F-FDG PET/CT radiomics-based machine learning analyses in breast tumors
| Authors | Years | Tumor type | Aim | Sample size | Constructed ML models | Core ML algorithm | Best ML model | Validation | Resultsa |
|---|---|---|---|---|---|---|---|---|---|
| Differentiating benign from malignant tumors and predicting tumor characteristics or stage | |||||||||
| Eifer et al. [92] | 2022 | Axillary LN | COVID-19 vaccine-associated lymphadenopathy vs. metastasis | n = 99 |
CT radiomics-based model PET radiomics-based model Combined model |
kNN | Combined model | Training and validation cohorts | AUC: 0.98 |
| Chen et al. [93] | 2022 | Breast cancer | HER2 status | n = 217 |
CT radiomics-based model PET radiomics-based model PET/CTconcat radiomics-based model PET/CTmean radiomics-based model |
XGB | PET/CTmean radiomics model | Training and validation cohorts | AUC: 0.760 |
| Song [94] | 2021 | Breast cancer | LNM | n = 100 | PET radiomics-based model alone | XGB | – | Training and validation cohorts | AUC: 0890 |
| Krajnc et al. [95] | 2021 | Breast cancer | Triple negative hormone status | n = 170 | Combined clinical + CT radiomics-based + PET radiomics-based model alone | Ensemble ML algorithm | – |
Internal validation (cross-validation) |
AUC: 0.82 |
| Ou et al. [96] | 2020 | Breast tumor | Breast cancer vs. malignant lymphoma | n = 44 |
SUV model CT radiomics-based model PET radiomics-based model Combined clinical + PET radiomics-based model Combined clinical + CT radiomics-based model |
LASSO + LDA | Combined clinical and PET radiomics model | Training and validation cohorts | AUC: 0.806 |
| Predicting treatment response or survival | |||||||||
| Li et al. [100] | 2020 | Breast cancer | pCR after NAC | n = 100 |
CT radiomics-based model PET radiomics-based model Combined age + CT radiomics-based + PET radiomics-based model |
RF | Combined model | Training and validation cohorts | Accuracy: 0.80 |
| Gómez et al. [101] | 2022 | Metastatic breast cancer | Metabolic response after treatment | n = 48 | Combined clinical + CT radiomics-based + PET radiomics-based model alone | LASSO + SVM | – | Training and validation cohorts | AUC: 0.82 |
AUC area under the receiver operating characteristic curve, HER2 human epidermal growth factor receptor, kNN k-nearest neighbors, LASSO least absolute shrinkage and selection operator algorithm, LDA linear discriminant analysis, LN lymph node, LNM lymph node metastasis, ML machine learning, NAC neoadjuvant chemotherapy, pCR pathological complete response, RF random forest, SVM support vector machine, XGB gradient tree boosting
aPerformance only presents the result of the best machine learning model