Table 7.
Anatomic site | Object or task | Network input | Network architecture | Dataset (train/test) |
---|---|---|---|---|
Bladder | Treatment response assessment268 | CT | CifarNet | 82/41 patients |
Brain | Glioblastoma multiforme treatment options and survival prediction218 | MRI | Custom | 75/37 patients |
Assessment of treatment effect in acute ischemic stroke269 | MRI | CNN based on SegNet | 158/29 patients | |
Breast | Response to neoadjuvant chemotherapy270 | MRI | Pretrained VGGNet followed by LDA | 561 examinations from 64 subjects LOO CV |
Response to neoadjuvant chemotherapy271 | MRI | Custom | 133/33 patients | |
Segmentation of clinical target volume272 | CT | Deep dilated residual network | 800 patients fivefold CV | |
Cancer cell lines | Prediction of drug effectiveness in cancer cell lines273 | Multiple omics data from cancer cells (gene expression data, copy number variation data, mutation data, and cell line annotations) | Deep autoencoder | 520/104 cell lines |
Head and Neck | Organ segmentation274 | CT | U‐Net based with shape retention model | 22/10 scans |
Kidney | Renal segmentation275 | CT | Custom | 89/24 patients |
Early detection of acute renal transplant rejection276 | DWI‐MRI | Stacked autoencoders | 100 patients fourfold, tenfold and LOO CV | |
Liver | Hepatobiliary toxicity prediction after liver SBRT277 | CT and patient demographics, clinical information | Custom CNN trained on other organs, fine‐tuned on liver SBRT | 125 patients 20‐fold CV |
Lung | Estimation of dose protocols in Radiotherapy278 | FDG‐PET/CT, clinical, genetic, imaging radiomics features, tumor and lung dosimetric variables, treatment plans | Deep Q‐Network | 114 real train/4000 synthesized test cases |
Dynamic tracking during therapy279 | DRRs from 4D CT | DenseNet | 1/9 volumes | |
Prostate | Prediction of dose from patient image contours280 | IMRT | U‐net | 80/8 patients |
Prediction of dosimetric eligibility of prostate cancer patients undergoing IMRT281 | CT | Fine‐tuned AlexNet | 60 patients fivefold CV | |
Pelvis | Generating synthetic CTs from MR‐only radiotherapy282 | MRI | CGAN | 123/59 patients |
Assessment of toxicity to normal organs and tissue283 | Rectum surface dose maps | Fine‐tuned VGG‐16 | 42 patients tenfold and LOO CV | |
Rectum | Segmentation of rectal tumors on T2‐MRI and clinical target volume segmentation on CT272 | T2‐MRI or CT | Novel CNN involving cascaded atrous convolution and spatial pyramid pooling | 70 T2‐MR and 100 CT fivefold CV |
Prediction of pathologic complete response after chemoradiation284 | CT | DNN classifier custom estimator | 95 patients fivefold CV |
IMRT, intensity‐modulated radiation therapy; SBRT, stereotactic body radiotherapy; DWI, diffusion‐weighted MRI; DRR, digitally reconstructed radiographs; LDA, linear discriminant analysis; LOO, leave‐one‐out; CV, cross‐validation.