Fully Automatic Volume Measurement of the Spleen at CT Using Deep Learning

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Accurate spleen segmentation is crucial for diagnosing various medical conditions.
• Manual segmentation is time-consuming and prone to inter-observer variability.
• Automated methods are needed to improve efficiency and consistency in clinical practice.

Purpose of the Study:

• To develop and validate a fully automated deep learning algorithm for spleen segmentation on CT scans.
• To assess the performance of the algorithm in a large patient cohort.
• To evaluate the algorithm's utility in detecting splenic volume changes.

Main Methods:

• A three-dimensional deep learning network was developed for spleen segmentation using thorax-abdomen CT scans.
• The algorithm was trained on 400 scans and tested on 50 scans from patients undergoing oncologic treatment.
• Performance was evaluated using Dice scores and compared against a splenic index equation and expert radiologists.

Main Results:

• The automated algorithm achieved a high Dice score (0.962), comparable to an independent observer (0.964).
• Radiologist agreement with reference standard for volume change detection improved from 81% to 92% when aided by the algorithm.
• The algorithm demonstrated robust performance on a large dataset.

Conclusions:

• Deep learning-based spleen segmentation is accurate and reliable on CT scans.
• The automated algorithm can assist radiologists in identifying abnormal splenic volumes and changes.
• This technology holds potential for improving diagnostic accuracy and workflow efficiency in radiology.