# SECONDGRAM: Self-conditioned diffusion with gradient manipulation for longitudinal MRI imputation

## TL;DR

SECONDGRAM is a new AI method that generates missing follow-up MRI scans, helping doctors study diseases like Alzheimer's over time using limited data.

## Contribution

Introduces self-conditioned diffusion with gradient manipulation to generate realistic longitudinal MRI data from scarce datasets.

## Key findings

- SECONDGRAM outperforms existing methods in modeling MRI patterns and downstream diagnostic accuracy.
- Self-conditioning enables effective use of unpaired MRI data to enrich limited datasets.
- Gradient manipulation improves model stability and prevents overfitting in low-data scenarios.

## Abstract

While individual MRI snapshots provide valuable insights, the longitudinal progression in repeated MRIs often holds more significant diagnostic and prognostic value. However, a scarcity of longitudinal datasets, comprising paired initial and follow-up scans, hinders the application of machine learning for crucial sequential tasks. We address this gap by proposing self-conditioned diffusion with gradient manipulation (SECONDGRAM) to generate absent follow-up imaging features, enabling predictions of MRI developments over time and enriching limited datasets through imputation. SECONDGRAM builds on neural diffusion models and introduces two key contributions: self-conditioned learning to leverage much larger, unlinked datasets and gradient manipulation to combat instability and overfitting in a low-data setting. We evaluate SECONDGRAM on the UK Biobank dataset and show that it not only models MRI patterns better than existing baselines but also enhances training datasets to achieve better downstream results over naive approaches.

•SECONDGRAM accurately imputes missing longitudinal MRI data•Self-conditioning effectively uses abundant unpaired MRI data•Gradient manipulation robustly prevents model overfitting•SECONDGRAM-imputed data boost downstream diagnostic accuracy

SECONDGRAM accurately imputes missing longitudinal MRI data

Self-conditioning effectively uses abundant unpaired MRI data

Gradient manipulation robustly prevents model overfitting

SECONDGRAM-imputed data boost downstream diagnostic accuracy

Longitudinal MRI is a type of medical imaging study where MRI scans are taken several times over a period of time to track changes. These studies play a pivotal role in understanding neurodegenerative diseases like Alzheimer’s disease and Parkinson’s disease, yet the limited availability of paired longitudinal imaging datasets significantly restricts advanced machine learning applications. Our research introduces SECONDGRAM, a robust framework utilizing neural diffusion models enhanced with self-conditioned learning and gradient manipulation techniques. These models are computer algorithms that simulate how information changes. SECONDGRAM addresses data scarcity by generating realistic follow-up MRI imaging features, thereby enriching limited datasets. This methodological advancement not only improves the accuracy and realism of imputed imaging features but also significantly boosts the performance of predictive models in critical downstream tasks. The implications extend beyond healthcare, presenting a versatile solution for data augmentation in numerous fields struggling with longitudinal data constraints, thus enhancing decision-making processes in precision medicine.

Longitudinal MRI datasets essential for neurodegenerative disease research are scarce, limiting advanced machine learning approaches. The authors address this challenge with a neural diffusion-based framework, SECONDGRAM, that can generate realistic follow-up MRI features. Leveraging self-conditioned learning and robust gradient manipulation, SECONDGRAM significantly enhances data quality and improves predictive performance for medical diagnoses, offering broad applicability in precision medicine and beyond.