In the past decade, there has been notable progress in machine learning (ML) and mathematical models aimed at enhancing healthcare. These advancements offer the potential to improve patient care, alleviate healthcare burdens, streamline processes, and empower individuals. Particularly, there has been rapid growth in utilizing computer-aided design (CAD), especially in areas like computer vision (CV), image analysis, and computational modeling, to advance healthcare. These ML methods and computational models can effectively handle large datasets to mitigate racial and socioeconomic biases and make timely disease-related decisions compared to experienced physicians. However, it's crucial to design these models meticulously to ensure reliability, robustness, and accuracy. This dissertation explores the integration of computational modeling and ML methods in healthcare advancement, with a specific focus on medical image analysis and model validation in kidney assessment, AMD detection, and pulse oximetry. In the domain of kidney assessment for transplant viability, the study introduces a novel deep learning architecture called Residual-Attention-UNET for the quantification of optical coherence tomography (OCT) kidney images. This model is shown to outperform currently used deep learning models in OCT kidney research (e.g. UNET). Utilizing OCT images, the introduced model showcases unparalleled performance in automatically segmenting proximal convoluted tubules (PCTs). Notably, this study demonstrates that the developed Residual-Attention-UNET model can reduce training time (∼2 times), improving IOU (∼2 times), dice scores (∼1.5 times), and leading to a faster convergence compared to the UNET model which can offer timely manner predictions on the donated kidney. Shifting to ophthalmology, the study addresses the complexities of diagnosing age-related macular degeneration (AMD) through a comprehensive strategy. This includes refining 11 retinal boundaries and utilizing deep ensemble learning methods to diagnose AMD in the early stages of its progress. To do so, the study proposes a graph-cut algorithm integrated with cubic spline techniques, to automate the delineation of up to 11 retinal boundaries in OCT retina images. The results show that the segmentation error in our layer refinement approach was significantly lower than in OCT Explorer (1.7% vs. 7.8%). Additionally, this research aimed at identifying AMD-related biomarkers such as retinal layer thicknesses, curvature, and radius of retinal pigment epithelium (RPE) layer, gray level contrast of layers, intensity-based layer distribution, and magnitude and phase of RPE which can help to understand the disease progress and improve diagnosis. The study shows that by employing a deep ensemble approach, the developed model substantially reduces segmentation errors compared to traditional methods, leading to exceptional diagnostic precision in detecting AMD. In the final study, the dissertation extensively investigates employing computational modeling in dental tissue and finger pulse oximetry, aiming to identify optimal testing practices for modeling diffuse optical spectroscopy (DOS) systems utilized in dentistry, respiratory conditions, and surgery. It advocates for the adoption of computational modeling, particularly Monte Carlo (MC) simulation, to acquire radiation-free 3D data from dental optical devices. Utilizing GPU-based MC methods, the research demonstrates that the system's ability to detect pulp signals is greater in the transmittance mode than in the reflectance mode, for both 633 nm and 1310 nm wavelengths. Moreover, in finger pulse oximetry modeling the objective is to uncover fundamental mechanisms that could inform the assessment of clinical devices and technological advancements. Furthermore, the study investigates the impact of sensor or bandage reflectivity on sensitivity toward the epidermal region. Simulation outcomes highlight a notable impact of skin pigmentation on transmitted signals, with modulation ratios at 660 nm and 940 nm demonstrating an inverse relationship with melanin fraction, particularly prominent at 660 nm (20-25%) compared to 940 nm (6-12%). These modulation ratios and the degree of pigmentation dependence are found to vary with arterial oxygen saturation (SaO₂)levels. The research outlined in this thesis explores medical image analysis and model validation to advance healthcare, with a specific emphasis on kidney assessment, AMD detection, and pulse oximetry. It presents groundbreaking findings, including the identification of novel disease patterns such as distinct morphometric features in kidney tissue indicative of transplant viability, aiming to improve the quality of life for patients suffering from end-stage renal diseases and abbreviate the wait time for receiving a transplanted kidney in timely manner and improving the automatic kidney assessment by at least 1.5 times compared with the conventional prediction models such as UNET, utilizing a fast, reliable, and accurate deep learning technique and advanced image analysis methods like contrast limited adaptive histogram equalization. The proposed Residual-Attention-UNET could capture more feature complexity, an enhanced performance, and expedite predictions on transplanted kidneys by 2 times compared to regular UNET models utilized by other researchers. Additionally, in AMD diagnosis, the thesis enhances ophthalmologists' decision-making processes and significantly diminishes diagnostic errors by 1.7% compared to 7.8% with traditional methods or human graders, by probing previously unidentified biomarkers such as retinal layer thicknesses and curvature of the retinal pigment epithelium (RPE) associated with AMD. As pulse oximeters play an increasingly vital role in diagnostics, particularly in detecting hypoxemia during the COVID-19 pandemic, recent research underscores errors in off-label applications, particularly concerning disparities in diagnostic accuracy based on factors like race or skin pigmentation. Notably, the positive bias in SpO₂ measurements, particularly pronounced in the critical range of 85-90%, raises significant concerns for devices designated for SpO₂ measurement. To investigate whether the observed disparities can be attributed to variations in the volume fraction of epidermal melanin MC modeling was employed. Ultimately, these findings hold the potential to revolutionize decision-making processes in clinical settings by enhancing our comprehension of medical conditions, refining diagnostic accuracy, and guiding personalized treatment strategies.