Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Automated classification of multi-parametric body MRI series

Boah Kim, Tejas Sudharshan Mathai, Kimberly Helm, Ronald M. Summers

TL;DR

This work tackles the problem of inconsistent DICOM header metadata that hinder automatic organization of mpMRI series by proposing a 3D DenseNet-121-based classifier for eight mpMRI series spanning chest, abdomen, and pelvis. Using five-fold cross-validation on 1,363 training studies and evaluation on 313 held-out studies across three Siemens scanners, the DenseNet-121 ensemble achieved high performance with average precision, sensitivity, and F1 around 96.6% and specificity of 99.6%. To date, this is the first study to classify mpMRI series at the chest–abdomen–pelvis level, offering a robust, automated pathway to optimize hanging protocols in modern radiology. The approach has potential to reduce radiologist workload and improve consistency, though future work should broaden vendor diversity and incorporate DICOM header metadata to enhance generalizability.

Abstract

Multi-parametric MRI (mpMRI) studies are widely available in clinical practice for the diagnosis of various diseases. As the volume of mpMRI exams increases yearly, there are concomitant inaccuracies that exist within the DICOM header fields of these exams. This precludes the use of the header information for the arrangement of the different series as part of the radiologist's hanging protocol, and clinician oversight is needed for correction. In this pilot work, we propose an automated framework to classify the type of 8 different series in mpMRI studies. We used 1,363 studies acquired by three Siemens scanners to train a DenseNet-121 model with 5-fold cross-validation. Then, we evaluated the performance of the DenseNet-121 ensemble on a held-out test set of 313 mpMRI studies. Our method achieved an average precision of 96.6%, sensitivity of 96.6%, specificity of 99.6%, and F1 score of 96.6% for the MRI series classification task. To the best of our knowledge, we are the first to develop a method to classify the series type in mpMRI studies acquired at the level of the chest, abdomen, and pelvis. Our method has the capability for robust automation of hanging protocols in modern radiology practice.

Automated classification of multi-parametric body MRI series

TL;DR

This work tackles the problem of inconsistent DICOM header metadata that hinder automatic organization of mpMRI series by proposing a 3D DenseNet-121-based classifier for eight mpMRI series spanning chest, abdomen, and pelvis. Using five-fold cross-validation on 1,363 training studies and evaluation on 313 held-out studies across three Siemens scanners, the DenseNet-121 ensemble achieved high performance with average precision, sensitivity, and F1 around 96.6% and specificity of 99.6%. To date, this is the first study to classify mpMRI series at the chest–abdomen–pelvis level, offering a robust, automated pathway to optimize hanging protocols in modern radiology. The approach has potential to reduce radiologist workload and improve consistency, though future work should broaden vendor diversity and incorporate DICOM header metadata to enhance generalizability.

Abstract

Multi-parametric MRI (mpMRI) studies are widely available in clinical practice for the diagnosis of various diseases. As the volume of mpMRI exams increases yearly, there are concomitant inaccuracies that exist within the DICOM header fields of these exams. This precludes the use of the header information for the arrangement of the different series as part of the radiologist's hanging protocol, and clinician oversight is needed for correction. In this pilot work, we propose an automated framework to classify the type of 8 different series in mpMRI studies. We used 1,363 studies acquired by three Siemens scanners to train a DenseNet-121 model with 5-fold cross-validation. Then, we evaluated the performance of the DenseNet-121 ensemble on a held-out test set of 313 mpMRI studies. Our method achieved an average precision of 96.6%, sensitivity of 96.6%, specificity of 99.6%, and F1 score of 96.6% for the MRI series classification task. To the best of our knowledge, we are the first to develop a method to classify the series type in mpMRI studies acquired at the level of the chest, abdomen, and pelvis. Our method has the capability for robust automation of hanging protocols in modern radiology practice.
Paper Structure (10 sections, 1 equation, 2 figures, 4 tables)

This paper contains 10 sections, 1 equation, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Data
  4. Model
  5. Experiments & Results
  6. Implementation details
  7. Results
  8. Discussion
  9. ACKNOWLEDGEMENTS
  10. Compliance with Ethical Standards

Figures (2)

  • Figure 1: The overall framework of our body mpMRI series classifier. Eight different MRI series {pre-contrast T1, T1-arterial phase, T1-venous phase, T1-delayed phase, T2-weighted, T2FS, DWI, and ADC} were used to train a 3D DenseNet-121 model with 5-fold cross-validation. At test time, the ensemble of models from each fold classified the series type of an input 3D MRI volume in the held-out test set.
  • Figure 2: Confusion matrix for our (a) ResNet-50 and (b) DenseNet-121 ensemble predictions on the test dataset (n = 313 studies). -p, -a, -v, and -d denote the pre-contrast, arterial, portal-venous, and delayed phase of T1-weighted imaging respectively.