Deep Learning Substitutes Gadolinium in Detecting Functional and Structural Brain Lesions with MRI

Abstract

While MRI contrast agents such as those based on Gadolinium are needed to enhance the detection of structural and functional brain lesions, there are rising concerns over their safety. Here, we hypothesize that a deep learning model, trained using quantitative steady-state contrast-enhanced MRI datasets in mice and humans, could generate contrast-equivalent information from a single non-contrast MRI scan. The model was first trained, optimized, and validated in mice. It was then transferred and adapted to human data, and we find that it can substitute Gadolinium-based contrast agents for detecting functional lesions caused by aging, Schizophrenia, or Alzheimer’s disease, and, for enhancing structural lesions caused by brain or breast tumors. Since derived from a commonly-acquired MRI, this framework has the potential for broad clinical utility and can be applied retrospectively to research scans across a host of diseases.

Deep-learning-based synthesis of post-contrast T1-weighted MRI for tumour response assessment in neuro-oncology: a multicentre, retrospective cohort study

“…26 Moreover, preliminary neuroimaging studies have also explored the technical feasibility of using dCNN for synthesising post-contrast T1-weighted MRI from the information included in pre-contrast MRI sequences. 7,8,27 In this study we went beyond the assessment of technical feasibility and specifically explored the clinical use of these synthetic post-contrast T1-weighted sequences by harnessing large-scale MRI data from several previous clinical trials in the field of neuro-oncology [9][10][11][12] alongside retrospective institutional data with more than 2000 patients from over 200 institutions. Our study used two popular dCNN architectures, namely an encoderdecoder architecture (U-Net) as a reference benchmark and a GAN architecture, which has gained substantial attention across multiple industries since its first description in 2014 by Goodfellow and colleagues 28 to generate synthetic instances of data that can pass for real data (eg, for image, video, and voice generation).…”

“…6 Hypothesis generating studies have also explored the potential of artificial neural networks for synthesising post-contrast MRI sequences from pre-contrast MRI sequences alone, thereby potentially bypassing the need of GBCA for particular applications such as brain tumour imaging. 7,8 Despite this interesting concept of synthesising postcontrast MRI sequences from pre-contrast MRI sequences, there are currently no independent large-scale studies on heterogeneous multi-institutional datasets and assessment of its diagnostic value for clinical decision making.…”

Deep Learning Identifies Neuroimaging Signatures of Alzheimer's Disease Using Structural and Synthesized Functional MRI Data

“…We utilized a pre-trained network called DeepContrast [13]. It is a deep learning approach proposed to perform quantitative structural-to-functional mapping, extracting the hemodynamic information from structural MRI.…”

“…ROC AUC: area under the receiver-operating characteristics curve. steps were previously described [13]. The entire pipeline is illustrated in the bottom path in Fig.…”

“…First, we organized a large-scale structural MRI dataset with over 2500 scans from age-matched AD and CN subjects, to control for aging confound effects in deep-learning-based AD classification. Second, we took advantage of our previous research [13,14] to synthesize high-resolution functional mappings from these structural MRI, resulting in a large-scale, paired, functional dataset. We used those structural and deep-learning-synthesized functional brain images to extend a prior state-of-the-art structural-based AD classification study [12].…”