Date on Master's Thesis/Doctoral Dissertation


Document Type

Doctoral Dissertation

Degree Name

Ph. D.


Computer Engineering and Computer Science

Degree Program

Computer Science and Engineering, PhD

Committee Chair

Elmaghraby, Adel

Committee Co-Chair (if applicable)

El-Baz, Ayman

Committee Member

Imam, Ibrahim

Committee Member

Park, Juw

Committee Member

Dwyer, Amy

Author's Keywords

Machine and deep learning; diagnosis; kidney diseases; kidney rejection; renal cancer


Machine learning (ML) and deep learning (DL) approaches have been used as indispensable tools in modern artificial intelligence-based computer-aided diagnostic (AIbased CAD) systems that can provide non-invasive, early, and accurate diagnosis of a given medical condition. These AI-based CAD systems have proven themselves to be reproducible and have the generalization ability to diagnose new unseen cases with several diseases and medical conditions in different organs (e.g., kidneys, prostate, brain, liver, lung, breast, and bladder). In this dissertation, we will focus on the role of such AI-based CAD systems in early diagnosis of two kidney diseases, namely: acute rejection (AR) post kidney transplantation and renal cancer (RC). A new renal computer-assisted diagnostic (Renal-CAD) system was developed to precisely diagnose AR post kidney transplantation at an early stage. The developed Renal-CAD system perform the following main steps: (1) auto-segmentation of the renal allograft from surrounding tissues from diffusion weighted magnetic resonance imaging (DW-MRI) and blood oxygen level-dependent MRI (BOLD-MRI), (2) extraction of image markers, namely: voxel-wise apparent diffusion coefficients (ADCs) are calculated from DW-MRI scans at 11 different low and high b-values and then represented as cumulative distribution functions (CDFs) and extraction of the transverse relaxation rate (R2*) values from the segmented kidneys using BOLD-MRI scans at different echotimes, (3) integration of multimodal image markers with the associated clinical biomarkers, serum creatinine (SCr) and creatinine clearance (CrCl), and (4) diagnosing renal allograft status as nonrejection (NR) or AR by utilizing these integrated biomarkers and the developed deep learning classification model built on stacked auto-encoders (SAEs). Using a leaveone- subject-out cross-validation approach along with SAEs on a total of 30 patients with transplanted kidney (AR = 10 and NR = 20), the Renal-CAD system demonstrated 93.3% accuracy, 90.0% sensitivity, and 95.0% specificity in differentiating AR from NR. Robustness of the Renal-CAD system was also confirmed by the area under the curve value of 0.92. Using a stratified 10-fold cross-validation approach, the Renal-CAD system demonstrated its reproduciblity and robustness with a diagnostic accuracy of 86.7%, sensitivity of 80.0%, specificity of 90.0%, and AUC of 0.88. In addition, a new renal cancer CAD (RC-CAD) system for precise diagnosis of RC at an early stage was developed, which incorporates the following main steps: (1) estimating the morphological features by applying a new parametric spherical harmonic technique, (2) extracting appearance-based features, namely: first order textural features are calculated and second order textural features are extracted after constructing the graylevel co-occurrence matrix (GLCM), (3) estimating the functional features by constructing wash-in/wash-out slopes to quantify the enhancement variations across different contrast enhanced computed tomography (CE-CT) phases, (4) integrating all the aforementioned features and modeling a two-stage multilayer perceptron artificial neural network (MLPANN) classifier to classify the renal tumor as benign or malignant and identify the malignancy subtype. On a total of 140 RC patients (malignant = 70 patients (ccRCC = 40 and nccRCC = 30) and benign angiomyolipoma tumors = 70), the developed RC-CAD system was validated using a leave-one-subject-out cross-validation approach. The developed RC-CAD system achieved a sensitivity of 95.3% ± 2.0%, a specificity of 99.9% ± 0.4%, and Dice similarity coefficient of 0.98 ± 0.01 in differentiating malignant from benign renal tumors, as well as an overall accuracy of 89.6% ± 5.0% in the sub-typing of RCC. The diagnostic abilities of the developed RC-CAD system were further validated using a randomly stratified 10-fold cross-validation approach. The results obtained using the proposed MLP-ANN classification model outperformed other machine learning classifiers (e.g., support vector machine, random forests, and relational functional gradient boosting) as well as other different approaches from the literature. In summary, machine and deep learning approaches have shown potential abilities to be utilized to build AI-based CAD systems. This is evidenced by the promising diagnostic performance obtained by both Renal-CAD and RC-CAD systems. For the Renal- CAD, the integration of functional markers extracted from multimodal MRIs with clinical biomarkers using SAEs classification model, potentially improved the final diagnostic results evidenced by high accuracy, sensitivity, and specificity. The developed Renal-CAD demonstrated high feasibility and efficacy for early, accurate, and non-invasive identification of AR. For the RC-CAD, integrating morphological, textural, and functional features extracted from CE-CT images using a MLP-ANN classification model eventually enhanced the final results in terms of accuracy, sensitivity, and specificity, making the proposed RC-CAD a reliable noninvasive diagnostic tool for RC. The early and accurate diagnosis of AR or RC will help physicians to provide early intervention with the appropriate treatment plan to prolong the life span of the diseased kidney, increase the survival chance of the patient, and thus improve the healthcare outcome in the U.S. and worldwide.