2024 ARRS ANNUAL MEETING - ABSTRACTS

RETURN TO ABSTRACT LISTING


E3255. AI-Based Prostate Segmentation on CT for Automatic Prostate Measurement: Ultrasonography as the Standard of Reference
Authors
  1. Roshan Fahimi; Massachusetts General Hospital
  2. Emiliano Garza Frias; Massachusetts General Hospital
  3. Lina Karout; Massachusetts General Hospital
  4. Parisa Kaviani; Massachusetts General Hospital
  5. Keith. J Dreyer; Massachusetts General Hospital
  6. Subba Digumarthy; Massachusetts General Hospital
  7. Mannudeep. K Kalra; Massachusetts General Hospital
Objective:
Measurement of prostate on abdomen-pelvis CT is both time-consuming and subjective. The purpose of our study was to assess performance of an AI-based organ segmentation algorithm (eXamine, Siemens Healthineers) for automatic prostate measurement with ultrasound as the standard of reference.

Materials and Methods:
Our study included 149 men (average age 66 ± 16 years) from nine academic, community, and cottage hospitals, who had a contrast-enhanced abdomen-pelvis CT and underwent a prostate ultrasound within 6 months of their CT. The eligible patients had either a normal prostate or benign prostate hyperplasia. Eligible patients were identified from a review of a commercial radiology report database search engine. Transverse CT images were anonymized and exported offline from the PACS. Exported datasets were processed with the eXamine prototype (Siemens Healthineers) for segmenting and measuring the prostate gland. The data were analyzed with multiple logistic regression, test with area under the curve for precision-recall curve analysis as the output (R statistical software).

Results:
Ultrasound-based prostate volumes for enlarged glands was 61 ± 26 mL, with corresponding volume of 67 ± 42 mL on CT with the organ segmentation and measurement. The AI prototype provided minimum, maximum, and average diameters as well as prostate volume. The CT-based automatic measurements had an AUC of 0.84 (<em>p</em> < 0.01) for differentiating normal and enlarged prostate glands. The maximum prostate diameter was the best parameter for differentiating normal and enlarged prostate.

Conclusion:
The AI-based organ segmentation and measurement prototype can reliably segment and measure prostate gland on CT with high accuracy. Clinical Relevance: AI-based organ segmentation and measurement tools can help radiologist automate measurements prone to subjective variations.