Machine Learning in Renal Failure using voice as biomarker

This thesis presents a novel approach to renal failure detection, proposing the use of voice analysis as a non-invasive biomarker. Renal failure, also known as kidney failure, is a condition affecting around 10% of the global adult population, occurs when the kidneys are unable to efficiently filter waste from the bloodstream, leading to fluid and toxic accumulation and other severe health complications. Current diagnostic methods rely on clinical assessments and laboratory tests, which are often time-consuming, resource intensive and stressful for the patient. This research explores an alternative, automated detection method by focusing on changes in vocal characteristics, hypothesizing that fluid retention can influence the voice in measurable ways. The primary goal of the study is to develop a machine learning model capable of detecting changes in patients’ voices that correspond to renal failure, particularly in those undergoing dialysis. Dialysis is the process of removing the toxin which the body is not bale to expel by removing the fluid accumulated in the body. By analyzing voice recordings from patients before and after dialysis, the study identifies patterns that correlate vocal changes with fluid removal during the treatment process. The dataset used includes voice recordings as in figure from 86 patients, collected over a period of 90 days. The analysis examines how different machine learning models—such as Support Vector Machines (SVM), Random Forest (RF), and Gradient Boosting (GB)—perform in classifying these changes. The study reveals several findings: • Support Vector Machines Outperform Other Models: Among the different models tested, SVM demonstrated the highest accuracy in detecting voice changes related to renal failure. This is particularly true for anuric patients, who experience more significant fluid retention. • Vocal Characteristics are Linked to Fluid Retention: The results show a significant correlation between fluid accumulation and changes in voice, supporting the hypothesis that voice can be a reliable biomarker for renal health monitoring. The presence of vocal changes, such as differences in pitch and tone, can signal fluid retention levels in patients. • Challenges in Model Generalization: While the models performed well in detecting voice changes for individual patients, generalizing these results across a broader population remains challenging. This suggests that future research should focus on refining the models to improve their applicability to diverse patient groups. This research introduces the potential for voice analysis to be used as a cost-effective, non-invasive tool in the clinical detection of renal failure. The use of voice as a biomarker not only reduces the reliance on resource-intensive tests but also provides an accessible means of monitoring patients over time.The study also highlights the challenges in developing models that can generalize well across a broader population. To address this, further research is needed to improve the robustness and accuracy of the models, particularly when applied to a more diverse set of patients. This thesis establishes a foundation for using vocal characteristics as a biomarker for renal failure, offering an alternative to traditional diagnostic methods. Using machine learning techniques, this approach could lead to the development of more efficient, patient friendly tools for monitoring renal failure.