This essay was initially featured in The Checkup, the regular biotechnology magazine of MIT Technology Review. To receive this content directly in your inbox every Thursday and access articles like this ahead of time, sign up here.
A recent discovery highlighted an innovative application of AI technology. Researchers have developed a smartphone app that can differentiate diseases by analyzing the sound of a patient’s cough.
While the method is groundbreaking, it is not infallible. The software had a 30% failure rate in detecting tuberculosis in individuals with the disease. Despite this limitation, the app’s utility lies in its cost-effectiveness compared to traditional TB diagnostic methods that involve collecting mucus samples. This aspect makes it particularly valuable as a monitoring tool in low-income regions, aiding in early case identification and disease prevention.
In a recent study, a team of researchers from the US and Kenya utilized their smartphone-based clinical tool to analyze audio recordings of coughs from 149 TB patients and 46 individuals with various respiratory conditions. Although the app cannot entirely replace conventional diagnostic techniques, it can serve as a supplementary screening tool. Early detection and treatment of active TB cases are crucial in curbing disease transmission.
The use of coughs and other bodily sounds as “acoustic biomarkers” to track health changes is a growing area of research. This approach, although not new, has gained significant traction in the past few decades, especially with the advancements in AI technology. Laryngologist Yael Bensoussan from the University of South Florida noted that AI enables rapid analysis of vast datasets.
The COVID-19 pandemic further popularized cough analysis. Numerous startups emerged, focusing on developing tools to diagnose diseases based on cough sounds. For instance, AudibleHealthAI introduced a mobile app for COVID-19 diagnosis, which is now under FDA review. The company is expanding its scope to include TB and influenza detection.
ResApp Health, an Australian company, shifted its focus to respiratory sound analysis post-COVID-19 outbreak. Their tool demonstrated a 92% accuracy rate in identifying positive COVID-19 cases based on cough sounds alone. Subsequently, Pfizer acquired ResApp for $179 million.
The potential of these applications as precise diagnostic tools remains uncertain. However, the ability to detect and analyze cough sounds can still offer valuable health insights. Cough sounds are easily recordable by smartphones, making them accessible for regular monitoring. Google Health’s latest Pixel phones feature cough and snore detection capabilities.
Bensoussan believes that cough-tracking applications could revolutionize clinical trials where cough analysis plays a vital role. She emphasized the challenges in accurately tracking coughs and highlighted the precision that mobile apps can provide in this regard.
Beyond cough analysis, Bensoussan is leading a $14 million NIH-funded project to create a comprehensive database of various sounds, including voices, coughs, and breathing patterns. This database aims to facilitate the development of diagnostic tools for a range of conditions, from cancers to mood disorders.
One limitation in this field is the prevalence of proprietary datasets, hindering research evaluation. Bensoussan and her team are working to make their database publicly accessible by June, promoting transparency and collaboration in sound-based health research.
As technology advances, we may witness more applications leveraging sound patterns for health monitoring and diagnostics. While the effectiveness of these tools is yet to be fully realized, ongoing developments hold promise for enhancing disease screening and diagnosis.
In a 2017 article by Emily Mullin, the potential of voice signals in identifying various health conditions, including PTSD, traumatic brain injury, mood disorders, and heart disease, was discussed.
Real-world implementation of AI tools may face challenges, as highlighted by Douglas Heaven’s account of Google Health’s diabetes-related screening tool deployment in Thailand. The balance between trusting AI recommendations and preserving patients’ experiences and clinical judgment was debated by Jessica Hamzelou in a previous issue of The Checkup, cautioning against over-reliance on AI in healthcare decision-making.
External sources:
- STAT: Evidence supporting the effectiveness of supervised consumption sites in saving lives, demonstrated through motion-sensor-equipped restrooms at a Boston clinic catering to homeless individuals on the “methadone mile.”
- STAT: Discussion on the limitations of pharmaceutical interventions in combating obesity, emphasizing the need for broader societal and economic changes to address the root causes of the issue.
- Wired: Concerns surrounding the potential of brain-computer interfaces to infringe on privacy and free will, prompting advocacy for “neurorights” and legislative safeguards against neurotechnology misuse.
