MediNet: transfer learning approach with MediNet medical visual database.

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Deep learning applications

Background:

• Deep learning requires large labeled datasets, which are scarce in medical research.
• Medical image analysis faces challenges in data collection and annotation.
• Existing deep learning models need effective pre-training strategies for medical tasks.

Purpose of the Study:

• To introduce MediNet, a novel 10-class medical imaging dataset.
• To evaluate the performance of various deep learning algorithms using transfer learning on the MediNet dataset.
• To enhance disease classification accuracy in medical imaging through pre-training with MediNet.

Main Methods:

• Development of the MediNet dataset comprising diverse medical imaging modalities (X-ray, CT, MRI, Ultrasound, Histopathology).
• Training and pre-training of seven deep learning algorithms (AlexNet, VGG19-BN, Inception V3, DenseNet 121, ResNet 101, EfficientNet B0, Nested-LSTM + CNN) on the MediNet dataset.
• Application of transfer learning using MediNet pre-trained models for classification tasks on Chest X-Ray, Covid-19, and Diabetic Retinopathy datasets.

Main Results:

• Transfer learning with MediNet significantly improved classification accuracy across multiple medical datasets.
• InceptionV3 achieved 98.71% accuracy on Chest X-Ray images after transfer learning, an increase from 94.84%.
• DenseNet121 and Nested-LSTM + CNN also showed improved performance on Covid-19 (92%) and Diabetic Retinopathy (81.52%) datasets, respectively, after MediNet pre-training.

Conclusions:

• MediNet is a valuable resource for training deep learning models in medical image analysis.
• Transfer learning using models pre-trained on MediNet enhances diagnostic performance for various medical conditions.
• The proposed approach demonstrates superior results in medical disease classification tasks.