Table 11 Comparative study table.

⁶

PPMR dataset

CDCM loss function

ResNet50

Recall—88.07, Precision – 71.86

⁷

Embryonic brain dataset

Deep features

SVM

Accuracy – 87.7

⁴

Training – Internal dataset. Testing – National Institute of Health (NIH) pediatric brain MRI database and the Developing Human Connectome Project (dHCP) database

Combining 2D and 3D CNN into an ensemble to predict myelin maturation age

3D CNN from⁴⁶ and EfficientNet-b0 as the 2D CNN

MAE Results:

Cross-validation set: 2D model – 1.53, 3D model – 2.06, Ensemble model – 1.63

Internal test set: 2D model – 1.43, 3D model – 2.55, Ensemble model – 1.77

External NIH dataset: 2D model – 2.26, 3D model – 2.27, Ensemble model – 1.22

External dHCP dataset: 2D model – 0.44, 3D model – 0.27, Ensemble model – 0.31

³²

Publicly available Brain Tumor MRI dataset. These images were classified into two classes: images with or without a tumor

Transfer learning to evaluate and compare multiple pre-trained deep learning models

VGG-16, Inception-v3, and ResNet50

Accuracies of VGG16 – 96, InceptionV3 – 78, ResNet50 – 95

Precision of VGG16 – 94, InceptionV3 – 75, ResNet50 – 92

Recall of VGG16 – 100, InceptionV3 – 70, ResNet50 – 89

F1-score of VGG16 – 98, InceptionV3 – 73, ResNet50 – 94

³⁴

Custom MRI dataset collected and augmented by Swati Kanchan from NIT Durgapur

Transfer learning and fine-tuning of MobileNet CNN; image resizing and normalization; GradCam for visual explanation

Fine-tuned MobileNet CNN

Validation Accuracy: 97.24; Test Accuracy: 97.86; Precision: 97.91; Recall: 97.86; F1-score: 97.86 for four class classification

Our

Approach

PPMR dataset⁶

Image processing pipeline using grayscale conversion, Min–Max normalization, histogram equalization, bilateral filtering, and Canny edge Detection

Modified DenseNet-201 and

MobileNetV2

Accuracies of DenseNet – 100, MobileNet – 99.8

Precision of DenseNet – 100, MobileNet – 99.6

Recall of DenseNet – 100,

MobileNet – 100