Table 5.
Application in assessing treatment response.
| Author | Year | Aim of study | NO. of Patients | MRI sequences | Machine learning/statistical method | outcomes |
|---|---|---|---|---|---|---|
| Galm (82) | 2018 | Predicting the P/R (progression or recurrence) | 78 (33 with P/R, 45 without P/R) | T1 | LR | Tumors with log-transformed mean pixel intensity above the median showed a 0.44 HR for recurrence or progression compared to lower intensity tumors. |
| Zhang (83) | 2020 | Predicting the P/R (progression or recurrence) | 50 (28 with P/R, 22 without P/R) | T2, T1CE | SVM | Radiomics analysis using preoperative CE T1WI and T2WI MRI could predict recurrence in NFPA and elevated radiomic scores correlated with reduced PFS times (p < 0.001). |
| Shen (84) | 2023 | Predicting residual tumor regrowth | 114 (70 with residual regrowth, 34 with no residual regrowth) | T1, T2, T1CE | LR | T1WI&T2WI outperformed other combinations or single sequences, and the integration of preoperative and postoperative images proved more effective than using them individually. |
| Zhang (87) | 2021 | Predicting postoperative visual field recovery | 131 (79 in the recovery group, 52 in the non-recovery group) | T2 | SVM, RF, LDA | Three radiomic models based on preoperative T2WI all showed good performance, each with an AUC over 0.75. |
| Zhang (88) | 2023 | Predicting postoperative visual field recovery | 130 (87 in the recovery group, 43 in the non-recovery group) | Preoperative and postoperative T2 | LASSO | Postoperative changes in the optic chiasm were not significant predictors of visual outcomes, but delta-radiomics of the optic chiasm have prognostic value for visual recovery. |
| Fan (95) | 2019 | predicts radiotherapeutic response in acromegaly | 57 (25 achieved remission, 32 did not) | T1, T1CE, T2 | SVM | The clinical-radiomics model showed good discrimination abilities, achieving an AUC of 0.96, surpassing that of any single clinical feature or standalone radiomics model. |
| Park (89) | 2021 | Predicting dopamine agonist response in prolactinoma | 177 (109 DA responders, 68 DA non-responders) | T2 | RF, light gradient boosting machine, ET, quadratic discrimination analysis, linear discrimination analysis and soft voting ensemble classifier | The ensemble classifier (AUC=0.81) performs better than any other individual machine learning classifier. Two second-order features demonstrated significant correlation with baseline PRL levels. |
| Galm (90) | 2020 | predicting response to somatostatin receptor ligands (SRLs) in acromegaly | 34 (17 SRL responders, 17 SRL non-responders) | T1 | LR | MRI texture of T1WI can predict normalization of IGF-I with SRL therapy. |
| Kocak (91) | 2019 | predicting response to somatostatin receptor ligands in acromegaly | 47 (24 SRL responders, 23 SRL non-responders) | T2 | KNN, C4.5 algorithm | Texture analysis based on KNN outperformed T2-weighted relative signal intensity, as well as immunohistochemical granulation pattern assessment, in predictive accuracy. |