MENU

Click here to

×

Are you sure ?

Yes, do it No, cancel

Development of Multi-Atlas-Based Prostatic Urethra Identification Method Using Machine Learning

H Takagi1*, N Kadoya2 , T Kajikawa2 , S Tanaka2 , Y Takayama2 , T Chiba2 , K Ito2 , S Dobashi1 , K Takeda1 , K Jingu2 , (1) Course of Radiological Technology, Health Sciences, Tohoku University Graduate School of Medicine, Sendai, 04, (2) Department of Radiation Oncology, Tohoku University Graduate School of Medicine, Sendai,

Presentations

(Sunday, 7/14/2019) 4:30 PM - 5:00 PM

Room: Exhibit Hall | Forum 9

Purpose: In the present study, we developed a multi-atlas based prostatic urethra identification method using machine learning (ML), and assessed its feasibility of replace use of urinary catheter.

Methods: We examined 100 patients with prostate cancer, treated with IMRT. All patients underwent temporary urinary catheter placement for identification and contouring of the prostatic urethra in pCT images (ground truth). We created an atlas-database by registering the planning CT images and the patients’ prostate, bladder, and urethra contours. We used four datasets from the database and performed structure-based deformable image registration (DIR). To determine the suitable four datasets we used support vector machine regression (SVR) and five feature descriptors to increase DIR accuracy. Finally, we combined the four deformed prostatic urethra contours into a single structure. In evaluation, 80 datasets were set to atlas database and predictions were made for remained 20 patient. To assess the accuracy of our method, we compared with three other methods including previous researches.

Results: The mean error in the entire prostatic urethra was 2.28 ± 1.15 mm, 2.63 ± 1.13 mm, 2.74 ± 1.32 mm, and 3.47 ± 1.19 mm for our method with ML, our method without ML, our method with RIR, and the centerline method respectively. Our method with ML showed the highest accuracy. In segmented prostate, mean error in the top 1/3 segment was highly improved, with results of 1.88 ± 1.03 mm, 2.40 ± 1.51 mm, 3.02 ± 1.76 mm, and 5.76 ± 3.09 mm respectively. Our method with ML resulted in a significantly smaller mean error in the top 1/3 segment compared with the other methods (p < 0.05).

Conclusion: We developed a multi-atlas based prostatic urethra identification method using ML. Our method had a great potential for replacing use of temporary indwelling of urinary catheter.

Keywords

Prostate Therapy, Segmentation

Taxonomy

IM/TH- image segmentation: CT

Contact Email