Deep learning for staging liver fibrosis on CT: a pilot study.

Area of Science:

• Radiology
• Artificial Intelligence
• Medical Imaging

Background:

• Liver fibrosis staging is crucial for patient management.
• Accurate staging traditionally relies on histopathology, which can be invasive.
• Non-invasive methods for fibrosis staging are highly desirable.

Purpose of the Study:

• To evaluate the efficacy of deep learning techniques in staging liver fibrosis using CT images.
• To develop and validate a deep convolutional neural network (DCNN) for liver fibrosis assessment.

Main Methods:

• A retrospective study utilized 496 contrast-enhanced CT examinations from 286 patients with available histopathology.
• A DCNN was trained on 396 images and validated on 100 images, correlating deep learning CT (F_DLCT) scores with fibrosis stage.
• Data augmentation techniques were employed to enhance model robustness.

Main Results:

• F_DLCT scores demonstrated a significant correlation with histopathological liver fibrosis staging (Spearman's rho = 0.48, p < 0.001).
• Areas under the ROC curves for diagnosing significant fibrosis (≥F2), advanced fibrosis (≥F3), and cirrhosis (F4) were 0.74, 0.76, and 0.73, respectively.
• The deep learning model achieved moderate performance in staging liver fibrosis.

Conclusions:

• Deep learning models applied to CT images show potential for staging liver fibrosis.
• The developed model achieved moderate accuracy, suggesting its utility as a non-invasive assessment tool.
• Further research and refinement are necessary for clinical implementation.