Deep Learning Reconstruction for DWIs by EPI and FASE Sequences for Head and Neck Tumors

Area of Science:

• Radiology
• Medical Imaging
• Oncology

Background:

• Diffusion-weighted imaging (DWI) using echo-planar imaging (EPI) often suffers from susceptibility artifacts.
• Fast advanced spin-echo (FASE) and deep learning reconstruction (DLR) show promise for enhancing DWI quality.

Purpose of the Study:

• To evaluate the impact of FASE and DLR on DWI quality, apparent diffusion coefficient (ADC) accuracy, and the differentiation of malignant from benign head and neck tumors.
• To compare FASE and EPI sequences with and without DLR in both phantom and patient studies.

Main Methods:

• In vitro phantom studies assessed ADC correlation with standard values using Spearman's analysis.
• In vivo studies on head and neck tumor patients compared signal-to-noise ratio (SNR) and ADC between EPI and FASE sequences, with and without DLR, using Tukey's test.

Main Results:

• In vitro: Excellent correlation (ρ ≥ 0.92) between measured and standard ADC values.
• In vivo: FASE with DLR showed significantly higher SNR than FASE without DLR (p=0.02).
• ADC values differed significantly between sequences with and without DLR for both benign and malignant tumors (p<0.05).

Conclusions:

• FASE sequence and DLR significantly improve DWI image quality and reduce distortions compared to EPI.
• These advancements do not substantially affect ADC measurements or the ability to distinguish malignant from benign head and neck tumors.