Table 1.
Summary of reviewed literature.
| Cancer type | References | No. of pts | Type of RT | Type of predicted toxicity | Features type | Classifier | Results* |
|---|---|---|---|---|---|---|---|
| Breast | (10) | 90 | RT | Dermatitis | R | RF | Acc = 0.87 (test) |
| (11) | 2277 | Moist desquamation, dermatitis, chest pain, fatigue | D, C | LR, RF, gradient boosting | 0.56–0.85 | ||
| (12) | 827 | RT | Telangiectasia | D, C | LASSO | ||
| Esophagus | (13) | 101 | IMRT or 3D-CRT | Pneumonitis | D, C | LR | Acc = 0.63 |
| Gyneco | (14) | 42 | EBRT+BRT | Rectal toxicity | D | SVM | 0.82–0.91 |
| (15) | 42 | EBRT+BRT | Rectal toxicity | D | CNN (transfer learning) | 1.29 | |
| (16) | 35 | BRT | Fistula formation | D, C | SVM | 1.30 | |
| H&N | (17) | 437 | RT (397) PT (40) | Toxicity (grade ≥3) | C | LR, RF, XGBoost | 0.63–0.65 |
| (18) | 2121 | RT | Unplanned hospitalizations, Feeding tube placement, Weight loss |
D, C | LR, gradient boosting, RF | 0.64–0.76 | |
| (19) | 153 | RT | Xerostomia | D, R, C | 6 ML algotithms | Best SVM and extra-trees 0.74–0.89 | |
| (20) | 86 | RT | Trismus | D | IBDM | Identification of a cluster of voxel related with toxicity | |
| (21) | 427 | RT | Xerostomia | D, C | LR, LASSO, RF | Best LR (0.70) | |
| (22) | 173 | RT | Acute dysphagia | D, C | SVM, RF | 0.82 | |
| (23) | 297 | IMRT | Xerostomia (grade ≥2) | D, C | LR | Model updating is beneficial |
|
| (24) | 134 | IMRT and PT | Esophagitis | R, D | LASSO | 0.75 | |
| (25) | 47 | 3D-CRT | Sensorineural hearing loss | R, C | Decision stump, Hoeffding | 76.08% accurarcy 75.9% precision | |
| (26) | 37 | IMRT | Parotid shrinkge Xerostomia |
D, C | Fuzzy logic Naïve Bayes |
Acc = 0.79–0.86 | |
| (27) | 249 | IMRT | Xerostomia, sticky saliva | R, D | Multivariate LR | 0.77 | |
| (28) | 351 | IMRT | Mucositis | D, C | LR, SVM, RF | 0.71 (RF) | |
| (29) | 1 (H&N) 1 (Prostate) |
IMRT | Xerostomia (H&N), Rectal bleeding (prostate) |
D | Decision tree, SVM | 0.42% MAE (H&N) 97% acc (prostate) | |
| Liver | (30) | 125 | SBRT | Hepatobiliary toxicity | D, C | CNN (transfer learning) | 1.25 |
| Lung | (31) | 110 | SBRT | LC, DFS, OS, and fibrosis | R | Cox regression | |
| (32) | 203 | IMRT or PT | Pneumonitis | C | RF | 1.06 | |
| (33) | 192 | IMRT and 3D-CRT | Radiation pneumonitis | R, D, C | LASSO | 0.68 | |
| (34) | 197 | SBRT | Chest wall syndrome | D, C | Descision tree RF |
n/a | |
| (4) | 3496 (lung+brain +H&N) |
RT | Classifiers comparison | D, C | Decision tree, RF, ANN, SVM, elastic net, logit-boost | Best: elastic net LR and RF | |
| (35) | 14 | SBRT | Lung injuries | R, D | LR | 0.64–0.78 | |
| (36) | 201 | SBRT | Pneumonitis | D, C | Decision trees, RF, RUSBoost | ||
| (37) | 115 | RT | Esophagitis | D, C | LASSO | 0.78 | |
| (38) | 54 | 3D-CRT | Pneumonitis | D, C | Bayesian network LR Single variable |
0.66–0.83 | |
| (39) | 748 | RT | Esophagitis | D, C | LR | 0.83 | |
| (40) | 219 | 3D-CRT | Pneumonitis | D, C | SVM | 1.16 | |
| (41) | 55 (H&N) 219+166 (Lung) |
3D-CRT | Xerostomia, Pneumonitis (166) Esophagitis (216) |
D, C | LR, SVM, ANN | Best: modified SVM | |
| (42) | 219 | RT | Radiation pneumonitis | D, C | Decision tree, ANN, SVM, self-organizing maps | 0.79 | |
| (43) | 234 | RT | Radiation pneumonitis | D, C | Decision tree | 0.72 | |
| (44) | 166 | EBRT | Esophagitis xerostomia |
D | LR | ||
| (45) | 142 | 3D-CRT | Pneumonitis | D | ANN | 0.61–0.85 | |
| Prostate | (46) | 64 | IMRT (52 pts), 3D-CRT (12 pts) | Urinary toxicity Gastro-intestinal toxicity |
R, D, C | LR | 0.65–0.77 |
| (47) | 33 | IMRT | Cystitis | R | LR | 0.62–0.75 | |
| (48) | 33 | IMRT | Rectal wall changes | R | LR | 0.46–0.81 | |
| (49) | 351 | RT | Rectal bleeding Fecal incontinence Urinary incontinence Nocturia |
R, D, C | LR | 0.58–0.73 | |
| (50) | 598 | RT | Late fecal incontinence | D, C | ANN | 0.78 | |
| (51) | 593 | RT | Rectal bleeding | D, C | ICA | 0.83, 0.80, 0.78 | |
| (52) | 324 | BRT+-EBRT | GU toxicity symptoms | D, C, G | RF | 0.7 | |
| (53) | 118 | EBRT, BRT | GI toxicities | D | LR | Identification of spatial constraint for toxicity reduction | |
| (54) | 368 | RT | Rectal bleeding, Erectile dysfunction |
C, G | RF, LR | 0.71 (rectal bleeding) 0.68 (erectile dysfunction) | |
| (55) | 79 | IMRT | Rectal toxicity (grade ≥2) | D, C | LR | 1.28 | |
| (56) | 754 | EBRT | Dysuria, hematuria, incontinence, frequency | D, C | LR, Elastic-net, SVM, RF, ANN, MARS |
Best: LR, MARS AUC = 0.65 |
|
| (57) | 99 | EBRT | Rectal bleeding | D | LDA, SVM, k-means, kNN, PCA, CP-DMA | Best: CP-DMA | |
| (58) | 261 | 3D-CRT | Rectal toxicity, rectal bleeding | D, C | RF NTCP, NTCP | 0.76, 0.66 | |
| (59) | 718 | RT | Rectal bleeding | LR, ANN | 0.655, 0.704 | ||
| (60) | 321 | RT | Acute bladder and rectal toxicity | D, C | ANN, SVM | 0.7 | |
| (61) | 119 | RT | Rectal bleeding Nocturia |
D | ANN | Sensitivity and specificity >55% |
3D-CRT, 3D conformal RT; Acc, accuracy; ANN, artificial neural network; AUC, area under the curve; BRT, brachytherapy; CNN, convolutional neural network; CP-DMA, canonical polyadic decomposition–deterministic multi-way analysis; DFS, disease free-survival; EBRT, external beam RT; GI, gastrointestinal; GU, genitourinary; H&N, head and neck; IBDM, image-based data mining; ICA, indipendent component analysis; IMRT, intensity-modulated RT; kNN, k-nearest neighbors; LASSO, Least Absolute Selection and Shrinkage Operator; LC, local control; LDA, linear discriminant analysis; LR, logistic regression; MAE, mean absolute error; MARS, multivariate adaptive regression splines; ML, machine learning; NTCP, normal tissue complication probability; n/a, not applicable; OS, overall survival; PCA, principal component analysis; pt, patient; PT, proton therapy; RF, random forest; RT, radiotherapy; RUSBoost, random under-sampling Boost; SBRT, stereotactic body RT; SVM, support vector machine. Features were classified as clinical (C), dosimetric (D), genomic (G), or radiomic (R).
If not specified, AUC values are reported.