Amr Mohamed et al. in January 2024 [24] proposed that the 3D Vision Transformer (ViT3D) with a 32x32x32 patch size and simple averaging ensemble outperformed other deep learning models in predicting MGMT biomarker status in glioblastoma, reaching a testing AUC of 0.6015. Exception outperformed EfficientNet-B3 and ResNet50, with a testing AUC of 0.61745.
Anton François et al. in January 2024 [25] explained that the new method for MRI image segmentation, Topological Data Analysis (TDA), has significant advantages over standard machine learning approaches. The process involves three steps: automatic thresholding to show the entire object, recognizing a distinct subset with predetermined topology, and deducing segmentation components. The final segmentation of the running model, displaying the Dice score of 0.94.
Mohannad Barakat et al. in December 2023[26] offered an innovative approach to multi-modal glioma segmentation that combines the Segment Anything Model (SAM) and a voting network. By fine-tuning SAM with bounding box-guided prompts (SAMBA), we may adapt the model to the intricacies of African data. To illustrate the approach's potential, trials on the BraTS-Africa dataset achieved excellent outcomes, with SAM achieving a Dice value of 86.6 for binary segmentation and 60.4 for multiclass segmentation.
Dapeng Cheng et al. in November 2023[27] emphasized EAV-UNet, a technology for accurately detecting lesion locations. Optimizing feature extraction, using automated segmentation algorithms to find anomalous regions, and strengthening the structure. The approach replaces the U-Net encoder with the VGG-19. To improve feature details, they added a CBAM (Channel and Spatial Attention Mechanism) module to the decoder. They added an edge identification part to the encoder to extract important edge characteristics from the stream. This strategy obtained an F1 score of 96.1%.
Numan Saeeda et al. in April 2023 [28] proposed that Deep learning algorithms were used to analyze brain MRI scans of tumors to detect the methylation status of the MGMT promoter. Researchers used deep learning algorithms and a large public MRI dataset of 585 people to predict the methylation status of the MGMT promoter in glioblastoma tumors. They evaluated these models using Grad-CAM, occlusion sensitivity, feature visualizations, and training loss landscapes. The highest AUC by fold and the largest mean AUC are detected. The mean AUC becomes saturated at 0.69.
Shahzad Ahmad Qureshi1[29] et al. in 2023 proposed a novel two-stage MGMT Promoter Methylation Prediction (MGMT-PMP) system that extracts latent features fused with radiomic features predicting the genetic subtype of glioblastoma. A novel fine-tuned deep learning architecture, namely Deep Learning Radiomic Feature Extraction (DLRFE) module, is proposed for latent feature extraction that fuses the quantitative knowledge to the spatial distribution and tumorous structural size by radiomic characteristics (GLCM, HOG, and LBP). The novice rejection method was shown to be highly effective in picking and isolating negative training cases from the original dataset. The fused feature vectors are then employed by k-NN and SVM classifiers for training and testing. The highest classification performance is (96.84 0.09%), (96.08 0.10) %, and (97.44 0.14) % for detecting MGMT methylation status in patients with glioblastoma.
In august 2022, Duyen Thi Do et al. [30] the radiomics features extracted from multimodal images from magnetic resonance imaging (MRI) had undergone a two stage feature selection method, including an extreme Gradient Boosting (XGBoost) feature selection model followed by a genetic algorithm (GA)-based wrapper model for extracting the most meaningful radiomics features for predictive purposes. According to the cross-validation results, the GA-based wrapper model performed well in predicting the MGMT methylation status in GBM, with a sensitivity of 0.894, specificity of 0.966, and accuracy of 0.925. Application of the extracted GBM radiomics features on a low-grade glioma (LGG) dataset also achieved a sensitivity 0.780, specificity 0.620, and accuracy 0.750, indicating the potential of the selected radiomics features to be applied more widely on both low- and high-grade gliomas.
After image preprocessing and feature extraction, Fei Zheng et al [31] in 2022 built and compared the performance of two types of machine-learning (ML) models in June 2022. The first type was set up using all MRI sequences (T1WI, T2WI, contrast enhancement (CE), FLAIR, DWI_b_high, DWI_b_low, and ADC), while the second type was established using single MRI sequences as described above. Results The Maximum Relevance and Minimum Redundancy technique was used to find seven radiomic features for the ML model based on all sequences. The predictive accuracy was 0.993 and 0.750 in the training and validation sets, respectively, and the area under curves (AUCs) were 1.000 and 0.754 in the two sets, respectively. For the ML model based on single sequences, the numbers of selected features were 8, 10, 10, 13,9, 7, and 6 for T1WI, T2WI, CE, FLAIR, DWI_b_high, DWI_b_low, and ADC, respectively, with predictive accuracies of 0.797 ~ 1.000 and 0.583 ~ 0.694 in the training and validation sets respectively, and the AUCs of 0.874 ~ 1.000 and 0.538 ~ 0.697 in the two sets, respectively. In the independent validation set, the T1WI-based model performed best, while the CE-based model performed worst.
In September 2022 Shingo Kihira et al. [32] developed a symmetric Deep Learning-based U-Net framework based on FLAIR's 512 _ 512 segmented maps as the ground truth mask. Their findings: The final group included 208 patients with an average _ standard deviation of age (years) of fifty-six _ fifteen and an M/F ratio of 130/78. The DSC of the generated mask was 0.93. Prediction of IDH-1 and MGMT status achieved AUCs of 0.88 and 0.62, respectively. Survival prediction of < 18 months proved an AUC of 0.75. Their deep learning-based framework can detect and segment gliomas with excellent performance for the prediction of IDH-1 biomarker status and survival.
In June 2022 Sixuan Chen et al [33], trained a residual network (ResNet) to give a binary prediction of MGMT promoter methylation status. Instead of using images as an input, as in existing research, our research extracted radiomics features from a selected region of interest (ROI) in different modalities of MR images and used them as the input of the model. Based on the ROI of the whole tumor, the predictive capacity of the T1CE and ADC model achieved the highest AUC value of 0.85. Based on the ROI of the tumor core, the T1CE and ADC model gained the most elevated AUC value of 0.90. After comparison, the T1CE combined with the ADC model based on the ROI of the tumor core showed the best performance, with the highest average accuracy (0.91) and AUC (0.90) among all models. The deep learning method using MRI radiomics has excellent diagnostic performance with a high accuracy in predicting MGMT promoter methylation in diffuse gliomas.