|
| Publication | Type(s) of cancer | Type of data | Methods | Performance |
|
| [66] | Breast cancer | Ultrasound images, 2 datasets (A & B) | LeNet, U-Net, AlexNet | (on dataset A) and (on dataset B) |
| [67] | Breast lesions | Two custom datasets | FCN-AlexNet, FCN-32s, FCN-16s, and FCN-8s | Dice score of 0.7626 (FCN-16s) |
| [68] | Breast cancer | Camelyon16 | DL algorithm | Slide-level AUC of 99% |
| [69] | Breast cancer | International Conference on Pattern Recognition 2012, MITOS-ATYPIA-14, Tumor Proliferation Assessment Challenge 2016 | Faster region CNN, ResNET-50, DenseNet-201 | 0.691 -measure for the MITOS-ATYPIA-14 dataset |
| [70] | Multiple | Custom histopathology image dataset | Multilayer perceptron, logistic modal tree, sequential minimal optimization, Naïve Bayes, random forest, rotation forest, J-Rip, and PART algorithms | Rotation forest algorithm achieved an accuracy of 85.7% for binary classification between cancerous and noncancerous cells |
| [71] | Breast cancer | ICPR 2014 mitosis dataset, TUPAC 2016 mitotic cell dataset | Modified regional CNN | , , on TUPAC 2016 dataset |
| [72] | Breast cancer | Custom dynamic contrast-enhanced MRI dataset | 3D deep CNN architecture | 83.7% accuracy, 90.8% sensitivity, 69.3% specificity, AUC of 0.859, overall dice distance of |
| [73] | Breast cancer | INbreast database | Different DL methods | Accuracy of 98.96%, MCC of 97.62%, -score of 99.24%, Jaccard similarity coefficient of 86.37% |
| [74] | Lung nodules | LUNA16, LIDC-IDRI | Two deep 3D customized mixed link network encoder-decoder architectures | Accuracy of 94.17% |
| [75] | Lung cancer | LIDC-IDRI dataset, Kaggle data science bowl challenge dataset | 3D CNN architectures | Dice coefficient for LIDC-IDRI of 0.40, with 0.25 precision and 0.93 recall |
| [76] | Bladder cancer | Custom datasets | DL algorithm | Per-frame sensitivity and specificity were 90.9% and 98.6% |
| [77] | Full body | PET images | DL-based approach | Dice score of |
| [78] | Brain metastases | Custom MRI dataset | DL-based approach (faster region-based CNN model) | 96% sensitivity, |
| [79] | Thyroid nodules | Two custom datasets of ultrasound images | You only look once v3 dense multireceptive field CNN | and 95.23 |
| [80] | Liver cancer | 225 CT scans of hemangioma, hepatocellular carcinoma, and metastatic carcinoma | Watershed segmentation, Gaussian mixture model (GMM), and deep neural network | Dice score of 0.9743, accuracy of 99.38% |
| [81] | Liver cancer | KMC liver dataset, multiorgan Kumar dataset | DL model combining residual block, bottleneck block, and attention decoder | Jaccard index of 0.7206 on KMC liver dataset and 0.6888 on Kumar dataset |
| [82] | Colorectal cancer | MICCAI gland segmentation dataset, colorectal adenocarcinoma gland dataset | Modified U-Net-based architecture | Dice score of 0.929 on MICCAI gland segmentation dataset, 0.89 on the colorectal adenocarcinoma gland dataset |
| [83] | Colorectal cancer | Custom dataset of MRI images of 28 adenocarcinomas and 5 mucinous carcinomas | CNN architecture which is a combination of three CNN architectures | Dice score of 0.60, precision of 0.76, and recall of 0.55 |
| [84] | Ovarian cancer | Custom dataset of 127 patients and a total of 469 images | U-Net models | Dice score of 0.87, an average Pearson correlation of 0.90, and an average intraclass correlation of 0.89 |
| [85] | Gastric cancer | Custom dataset of 1208 healthy and 533 endoscopic images | Mask R-CNN algorithm | Average dice index of 71% |
| [86] | Gastric cancer | Custom dataset of 1340 pathological slices | Deeplab v3+ | Dice score of 0.9166 |
| [87] | Prostate cancer | MRI images from an online database | DL system combining four U-Net models | Overall average accuracy of 95.3% |
| [88] | Prostate cancer | Custom dataset of 1200 ultrasound images | DL method integrating mask R-CNN and Inception version 3 models | Dice score of 0.88, a precision of 76% on malignant and 75% on benign classes for the classification task using an Inception v3 architecture |
| [89] | Prostate cancer | Custom dataset of CT images of 556 cases | 2D U-Net model | Dice score of 0.85, 0.94, and 0.85 for prostate, bladder, and rectum, respectively |
| [90] | Pancreatic cancer | Custom dataset of T1w MRI images of 40 subjects | CNN architecture | Dice score of 0.73 |
| [91] | Pancreatic cancer | Custom dataset of MRI images belonging to 73 patients | DL method using spiral transformation | Dice score of |
|
