Super-Resolution Deep Learning Framework for Enhanced ADPKD Monitoring via Multi-Sequence Longitudinal MRI

ABSTRACT Autosomal Dominant Polycystic Kidney Disease (ADPKD) is a prevalent genetic disorder affecting millions worldwide, characterized by cyst formation primarily within the kidneys, leading to organ enlargement and deteriorating function. This condition often extends to the liver, posing significant clinical challenges. Traditional qualitative analysis methods are insufficient for accurate disease assessment, prognosis, and treatment guidance. Deep Learning (DL) models have shown promise in estimating Total Kidney Volume (TKV) from MRI images but face limitations in detecting early-stage disease with small cysts. This study aims to transform qualitative analysis into quantitative assessment through multi-class cyst segmentation in ADPKD. Aim 1 focuses on developing a super-resolution DL model to generate high-resolution 3D MR volumes. Multiple imaging planes (axial, coronal, sagittal) will be incorporated to enhance 3D resolution for precise biomarker calculation. Additionally, a multi-class multi-sequence DL framework will be developed for ADPKD severity assessment, which involves creating segmentation and object detection models for different cyst classes, including simple, hemorrhagic, and exophytic cysts within the kidney, liver, and pancreas. Aim 2 aims to build a prognostic model for predicting estimated glomerular filtration rate (eGFR) decline and disease progression. This will involve developing an accurate predictive model based on biomarkers extracted from high-resolution MR images, with a particular focus on cyst class information, especially hemorrhagic cysts, to improve eGFR decline forecasts and dialysis potential predictions. Aim 3 integrates patients' longitudinal data to quantify the impact of temporal dynamics on disease progression over the next 10 years, by developing a multimodal predictive model that incorporates patients' prior MRI scans and historical data. This study aims to utilize DL models to shift from qualitative to quantitative assessment of ADPKD using multi-sequence high-resolution MR images, enabling precise measurement of cyst attributes, and advancing our understanding of disease progression and treatment response for better patient outcomes.