Automated Detection of Poor-Quality Scintigraphic Images Using Machine Learning

Area of Science:

• Medical Imaging
• Machine Learning
• Radiology

Background:

• Automated detection of poor-quality scintigraphic images is crucial for reliable diagnostic interpretation.
• Traditional methods for image quality assessment can be subjective and time-consuming.

Purpose of the Study:

• To develop and validate a machine learning algorithm for automated detection of poor-quality scintigraphic images.
• To assess the accuracy of the machine learning algorithm on 99mTc-MDP bone scan images.

Main Methods:

• Utilized principal component analysis (PCA) to extract 32 feature vectors from scintigraphic images.
• Employed a multivariate adaptive regression splines (MARS) machine learning algorithm for image classification.
• Trained and tested the model on datasets with a 60:40 split, using five-fold cross-validation and ROC analysis for optimal model selection.

Main Results:

• The machine learning model achieved an accuracy of 93.22% on the training dataset and 86.88% on the test dataset.
• Sensitivity and specificity for classifying poor-quality and good-quality images were high, reaching 80% and 93.7% respectively on the test set.
• The algorithm demonstrated robust performance in distinguishing between good-quality (low-speed acquisition) and poor-quality (high-speed acquisition) images.

Conclusions:

• Machine learning algorithms, specifically MARS with 32 PCs, can effectively classify scintigraphic image quality.
• The developed automated method offers a reliable and accurate approach for detecting poor-quality bone scan images.
• This technology has the potential to improve the efficiency and consistency of image quality control in nuclear medicine.