LMTTM-VMI: Linked Memory Token Turing Machine for 3D volumetric medical image classification

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Deep Learning

Background:

• Biomedical imaging is crucial for medical diagnosis and treatment.
• Integrating deep learning into medical imaging faces challenges with 3D volumetric data.
• Traditional methods struggle with spatial characteristics and memory efficiency in 3D images.

Purpose of the Study:

• To introduce a novel deep learning model, the Linked Memory Token Turing Machine (LMTTM), for improved 3D volumetric medical image classification.
• To enhance the processing of spatial dependencies and structural complexities in medical images.
• To improve memory utilization and model adaptability for complex medical image analysis.

Main Methods:

• The study proposes the Linked Memory Token Turing Machine (LMTTM) model.
• LMTTM utilizes an external linked memory module to record image features.
• The model is evaluated on six 3D volumetric medical image datasets from MedMNIST v2.

Main Results:

• The LMTTM model achieved an average accuracy (ACC) of 82.4% on MedMNIST v2 datasets.
• The proposed model attained state-of-the-art (SOTA) performance in 3D volumetric medical image classification.
• Ablation studies showed a 5.7% performance improvement over the predecessor TTM's Memory.

Conclusions:

• LMTTM effectively classifies 3D volumetric medical images by leveraging linked memory for feature accumulation and reasoning.
• The model demonstrates significant potential to assist healthcare professionals in diagnosis and treatment planning.
• The developed LMTTM offers enhanced capabilities for complex medical image analysis.