Automated Deep Learning Artificial Intelligence Tool for Spleen Segmentation on CT: Defining Volume-Based Thresholds...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Computational Pathology

Background:

• Traditional linear measurements for assessing splenomegaly are often inaccurate.
• Prior research demonstrated a deep learning artificial intelligence (AI) tool's capability for automated spleen segmentation and volume determination.

Purpose of the Study:

• To apply an automated deep learning AI tool to a large screening population.
• To establish reliable volume-based splenomegaly thresholds.
• To evaluate the performance of AI-derived volumetric measurements against traditional linear methods.

Main Methods:

• Retrospective analysis of 8901 patients in a primary screening sample and 104 patients in a secondary sample with end-stage liver disease.
• Utilized a deep learning AI tool for automated spleen segmentation and volume calculation.
• Derived weight-based volumetric thresholds for splenomegaly using regression analysis and validated against radiologist assessments.

Main Results:

• A weight-based volumetric threshold for splenomegaly was established: (3.01 × weight [kg]) + 127 mL, with a constant threshold of 503 mL for weights >125 kg.
• In the secondary sample, 84% of patients met the weight-based volume-defined splenomegaly threshold.
• AI-based volumetric assessment showed superior performance compared to linear measurements for splenomegaly detection.

Conclusions:

• A novel, weight-based volumetric threshold for splenomegaly was successfully derived using an automated AI tool.
• The AI tool demonstrates potential for facilitating large-scale, opportunistic screening for splenomegaly.
• This AI-driven approach offers a more accurate and efficient method for spleen volume assessment in clinical practice.