A prior knowledge guided deep learning based semi-automatic segmentation for complex anatomy on MRI.

Researchers

Journal

Modalities

Models

Abstract

Despite recent substantial improvement in autosegmentation using deep learning (DL) methods, a labor-intensive and time-consuming slice-by-slice manual editing is often needed particularly for complex anatomy (e.g., abdominal organs). This work aims to develop a fast, prior knowledge guided DL semi-automatic segmentation (DL-SAS) method for complex structures on abdomen MRIs.A novel application of using contours on an adjacent slice as prior knowledge informant in a 2D-UNet DL model to guide auto-segmentation for a subsequent slice was implemented for DL-SAS. A generalized, instead of organ-specific, DL-SAS model was trained and tested for abdominal organs on T2-weighted MRIs collected from 75 patients (65 for training and 10 for testing). The DL-SAS model performance was compared with three common auto-contouring methods (linear interpolation, rigid propagation, and a full 3D DL auto-segmentation model trained with the same training dataset) based on various quantitative metrics including Dice similarity coefficient (DSC) and ratio of acceptable slices (ROA) using paired t-tests.For the 10 testing cases, the DL-SAS model performed best with the slice interval (SI) of 1, resulting in average DSC = 0.93±0.02, 0.92±0.02, 0.91±0.02, 0.88±0.03, and 0.87±0.02 for large bowel, stomach, small bowel, duodenum, and pancreas, respectively. The performance decreased with increased SIs from the guidance slice. The DL-SAS method performed significantly better (p<0.05) than other three methods. The ROA values were in the range of [48%-66%] for all the organs with SI=1 for DL-SAS, higher than those for liner interpolation ([31%-57%] for SI=1), and DL auto-segmentation ([16%-51%]).The developed DL-SAS model can segment complex abdominal structures on MRI with high accuracy and efficiency and may be implemented as an interactive manual contouring tool or a contour editing tool in conjunction with a full auto-segmentation process, facilitating fast and accurate segmentation for MRI-guided online adaptive radiation therapy.Copyright © 2022. Published by Elsevier Inc.