Developing Novel Disease Screening Tools by Measuring Critical Biochemical and Physiological Signals

Abstract

Health screening tools that are reliable and accurate have successfully reduced the burden on expensive diagnostic testing along with providing the opportunity of quick, lower cost treatments to a larger community. Designing these tools require interdisciplinary efforts, and in this thesis, author explores development of screening tools for two different bacterial diseases in collaboration with clinicians, researchers, and engineers. Oral Caries and Pul- monary Tuberculosis (TB), bacterial diseases affecting millions across the world, face unique challenges in early disease screening. In the case of oral caries, quantitative screening tools are very limited while in TB screening, the lack of low-cost screening device leaves millions of TB cases undetected. This thesis presents novel sensing techniques based on measuring critical biomarkers, identified with the help of clinicians, that can provide rapid, accurate, quantitative, and non-invasive feedback. Particularly, to reduce gaps in dentistry, thesis fo- cuses on developing tools to measure acidity of oral biofilm. This acidity plays a vital role in the formation of oral caries, yet tools to measure it clinically are absent. O-pH, an optical pH monitoring device is presented that measures the acidity of dental plaque using non-invasive, opto-electronic sensors and provides quantitative feedback of oral health to dentists. The results from in vitro validation and in vivo study with 30 subjects indicate that the device is reliable and accurate. The spot based device is extended to mapping pH using images and a proof-of-concept device built with multimodal-scanning fiber (mm-SFE) is presented. This technology can enable trend based oral health monitoring using pH. In the next part of the thesis, the similar design process is applied to develop a screening tool for TB that focuses on cough as a biomarker. The production of cough in TB patients are primarily in response to changes in lung tissues, which the thesis hypothesizes can result in differences in cough sounds compared to other respiratory health issues. To validate, the author presents, TBscreen, a ResNet-18 model, based on scalogram images of passive coughs. TBscreen was trained and validated using a controlled dataset of passive cough sounds of subjects with TB and other respiratory health ailments. The analysis indicates the sounds from coughing contain disease biomarkers that increases with bacterial load and has the potential to enable large scale screening. Further, results indicate that passive or natural coughs differ from forced coughs, coughs produced on an external prompt, making it difficult to use forced coughs as a proxy for passive coughs. Through development of these two new sensing tools into human subject studies, the author argues that significant contributions have been made to the engineering and medical communities at large.