Related Experiment Video
Updated: Sep 3, 2025
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
DAFT: A universal module to interweave tabular data and 3D images in CNNs.
Tom Nuno Wolf1, Sebastian Pölsterl2, Christian Wachinger1
1The Lab for Artificial Intelligence in Medical Imaging (AI-Med), Department of Child and Adolescent Psychiatry, Ludwig-Maximilians-Universität, Nussbaumstraße 5, Munich 80336, Germany; Technical University of Munich, School of Medicine, Department of Radiology, Ismaninger Straße 22, Munich 81675, Germany.
This study introduces a new method, Dynamic Affine Feature Map Transform (DAFT), to combine 3D brain images and clinical data for Alzheimer's Disease (AD) diagnosis. DAFT significantly improves diagnostic accuracy by integrating diverse patient information.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Neurology
Background:
- Convolutional Neural Networks (CNNs) effectively use 3D image data for Alzheimer's Disease (AD) diagnosis.
- Established AD biomarkers exist as tabular data (e.g., demographics, genetics, CSF analysis).
- Integrating tabular clinical data into CNNs for improved AD diagnosis remains underexplored.
Purpose of the Study:
- To introduce a novel module, Dynamic Affine Feature Map Transform (DAFT), for integrating tabular data into CNNs.
- To enhance the diagnostic capabilities of CNNs by leveraging both 3D imaging and clinical information.
- To improve Alzheimer's Disease diagnosis and time-to-dementia prediction accuracy.
Main Methods:
- Developed DAFT, a general-purpose module for CNNs, to dynamically transform feature maps.
- Utilized an auxiliary neural network within DAFT to generate scaling factors and offsets based on image and tabular data.
- Applied DAFT to combine 3D image data with tabular clinical information for AD diagnosis and prediction tasks.
Main Results:
- DAFT achieved a mean balanced accuracy of 0.622 for AD diagnosis.
- DAFT achieved a mean c-index of 0.748 for time-to-dementia prediction.
- DAFT outperformed all baseline methods in both diagnostic and predictive tasks, demonstrating its effectiveness.
Conclusions:
- The DAFT module effectively integrates 3D imaging and tabular clinical data for improved AD diagnosis and prediction.
- DAFT offers a robust and generalizable approach for enhancing CNN performance by incorporating diverse data types.
- Further research and application of DAFT can advance AI-driven diagnostic tools in neurology.
