AI Model Outperforms Endocrinologists in Diagnosing Acromegaly via Voice Recognition
Recent advancements in artificial intelligence (AI) have demonstrated substantial potential in the medical field, specifically in diagnosing rare diseases like acromegaly. A groundbreaking study conducted at Umeå University in Sweden has revealed that an AI model using voice recognition technology was more effective than experienced endocrinologists at diagnosing acromegaly—a disorder characterized by an excess of growth hormone, typically resulting in enlarged features—in adults. This study was published in The Journal of Clinical Endocrinology & Metabolism.
The research team, led by Dr Per Dahlqvist and Dr Konstantina Vouzouneraki, focused on the unique characteristics of voice as a potential biomarker for acromegaly. Participants were requested to pronounce a sustained letter "A" and read a 160-word text. The researchers collected voice recordings from 151 adults diagnosed with acromegaly, who had a median age of 57, with 39% being women. An equal number of matched controls based on age and sex were also included.
A significant portion of the recordings, 76%, was used to train three machine learning models which were then stacked together. These models were assessed using a test set comprising 31 individuals with acromegaly and 31 controls. The AI model demonstrated a sensitivity of 71% and a specificity of 77%, with an area under the receiver operating curve (AUC) of 0.84, surpassing the endocrinologists' performance which had an AUC of 0.69.
Interestingly, the 12 endocrinologists participating in the study—experts in dealing with pituitary diseases—assessed 50 acromegaly patients and 50 controls by listening to voice samples that included a sustained "A" and four short sentences. Their sensitivity was reported at 41%, and specificity was 75%. Notably, their performance dropped to an AUC of 0.57 when only listening to the sustained "A" sound.
This novel research highlights the potential for integrating AI and voice recognition in medical diagnostics, particularly for acromegaly. The Voice Handicap Index, evaluating the psychosocial impact of voice disorders, revealed higher scores among those with acromegaly, indicating significant voice limitations.
Dr Dahlqvist and Vouzouneraki suggest that the success of this AI model in distinguishing voice characteristics associated with acromegaly could spearhead further development and application of AI in diagnosing other diseases. They highlight that although the potential is evident, further training and validation of these models are essential before they can be implemented in everyday clinical settings.
The study underscores the promise of digitalisation and machine learning, noting that collaboration across international borders could accelerate progress in diagnosing rare diseases more effectively. The potential use in primary care or high-risk groups, such as sleep apnea clinics, could provide a new avenue for early detection and management of acromegaly, potentially reducing the considerable diagnostic delays currently experienced.
Source: Noah Wire Services