Title:Transfer Learning of the ResNet-18 and DenseNet-121 Model Used to Diagnose
Intracranial Hemorrhage in CT Scanning
Volume: 28
Issue: 4
Author(s): Qi Zhou, Wenjie Zhu, Fuchen Li, Mingqing Yuan, Linfeng Zheng*Xu Liu*
Affiliation:
- Shanghai General Hospital, Shanghai Jiao Tong University, Shanghai, 200042,China
- Medical College of Guangxi University, Nanning, Guangxi,China
Keywords:
Deep learning, transfer learning, intracranial hemorrhage, ResNet, DenseNet diagnosis, CT scanning.
Abstract: Objective: The aim of the study was to verify the ability of the deep learning model to identify five
subtypes and normal images in non-contrast enhancement CT of intracranial hemorrhage.
Methods: A total of 351 patients (39 patients in the normal group, 312 patients in the intracranial hemorrhage
group) who underwent intracranial hemorrhage noncontrast enhanced CT were selected, obtaining 2768 images
in total (514 images for the normal group, 398 images for the epidural hemorrhage group, 501 images for the
subdural hemorrhage group, 497 images for the intraventricular hemorrhage group, 415 images for the cerebral
parenchymal hemorrhage group, and 443 images for the subarachnoid hemorrhage group). Based on the diagnostic
reports of two radiologists with more than 10 years of experience, the ResNet-18 and DenseNet-121
deep learning models were selected. Transfer learning was used. 80% of the data was used for training models,
10% was used for validating model performance against overfitting, and the last 10% was used for the final
evaluation of the model. Assessment indicators included accuracy, sensitivity, specificity, and AUC values.
Results: The overall accuracy of ResNet-18 and DenseNet-121 models was obtained as 89.64% and 82.5%, respectively.
The sensitivity and specificity of identifying five subtypes and normal images were above 0.80. The
sensitivity of the DenseNet-121 model to recognize intraventricular hemorrhage and cerebral parenchymal hemorrhage
was lower than 0.80, 0.73, and 0.76, respectively. The AUC values of the two deep learning models
were found to be above 0.9.
Conclusion: The deep learning model can accurately identify the five subtypes of intracranial hemorrhage and
normal images, and it can be used as a new tool for clinical diagnosis in the future.
