Medical image reconstruction is the process of reproducing an image of an object from the measurements produced from a scanner through some physical process. Deep image reconstruction, which is utilizing deep learning in image reconstruction, has surpassed what is possible compared to classical approaches. However, one requirement of deep learning is a large dataset that covers as many cases as possible and covers different demographics, disease conditions, hardware configurations, and acquisition schemes. Thus, a major limiting factor in deep learning in medical image reconstruction is the training data available. In this dissertation, I address the challenge of imperfect and small data in deep learning. This is addressed through the use of Bayesian deep learning. Bayesian deep learning allows a practitioner to estimate uncertainty to quantify possible sources of error and reduce the tendency of models to overfit. This is used in two application domains: simultaneous positron emission tomography and magnetic resonance imaging (PET/MRI) attenuation correction and in sub-Nyquist accelerated MRI. The use of uncertainty in PET/MRI attenuation correction has allowed the correction of intrinsic dataset errors due to bowel air movement, lack of bone signal, misregistration, and metal implant artifacts. The use of uncertainty in sub-Nyquist accelerated MRI has reduced the overfitting behavior of reconstruction models and allowed for higher possible acceleration factors.

Tumor grading is an important prognostic parameter for renal cell carcinoma (RCC). However, current grading schemes require an invasive surgical procedure, putting patients at risks including increased risk of hemorrhage, infection, renal failure, or death. Furthermore, low grade RCC is indolent with low mortality risk and may not require treatment. Therefore, a pre-operative and non-invasive assessment of malignancy grade may be beneficial and facilitate optimal timing of treatment. In recent years, deep learning-based image analysis has gained wide popularity in cancer prognosis and prediction. The goal of this study is to investigate the feasibility and performance of a deep-learning-based model for clear cell RCC grading prediction from contrast-enhance computed tomography (CECT). After institutional review board approval, an institutional pathology database was queried for all renal biopsies between December 2002 and October 2018. All included patients received a CT with a non-contrast and at least one post-contrast series. All patients have Fuhrman grade confirmation from surgical pathology, with CT scan prior to the procedure. Tumors were manually annotated by a radiologist on either the corticomedullary or nephrographic phase CECT. Rectangular regions of interest (ROI) were drawn on each slice throughout the tumor and used as inputs to the Deep CNN ResNet50. A binary label of low grade or high grade was assigned to each patient. Sensitivity, specificity, accuracy, and AUC were calculated based on a five-fold cross-validation. Preliminary results from a small subset of data suggests that a deep learning model can be used to predict clear cell RCC grading based on CT imaging prior to surgical procedures.

Purpose

Methods

Results

Conclusion