Umeå University's logo

Context: Substantial diagnostic delay in acromegaly contributes to increased morbidity and mortality. Screening attempts in high-risk groups have yielded few positive cases, underscoring the need for simple and precise prescreening methods.

Objective: Machine-learning analysis of facial images shows promise for acromegaly detection but requires validation in larger, well-characterized cohorts using robust machine-learning frameworks as performed in this study.

Methods: Facial images from different angles were collected via smartphone from 155 acromegaly patients (79% biochemically controlled) and 153 matched controls at all Swedish university hospitals. Six machine-learning models were trained to distinguish acromegaly from control images, including 3 deep neural networks pretrained on diverse image datasets (ImageNet models: ResNet50, InceptionV2, and DenseNet121) and 1 network pretrained specifically on facial images (FaRL). Model performance was compared to assessment by 12 experienced endocrinologists.

Results: The diagnostic accuracy of the FaRL-based model was superior to all ImageNet models and matched the accuracy of human experts (area under the receiver operating characteristic curve 0.89 for both) with similar specificity (0.87 vs 0.93) but higher sensitivity (0.82 vs 0.66). Classification agreement between the best machine-learning model (FaRL) and human experts was 86% for true negatives and 60% for true positives. Machine-learning models and human experts both showed greater sensitivity in identifying acromegaly in male patients.

Conclusion: A deep learning model pretrained on facial features (FaRL) can detect acromegaly from standard photographs with accuracy comparable to that of expert endocrinologists. This supports the feasibility of face analysis as a screening tool for acromegaly.

Context: There is a considerable diagnostic delay in acromegaly, contributing to increased morbidity. Voice changes due to orofacial and laryngeal changes are common in acromegaly.

Objective: Our aim was to explore the use of digital voice analysis as a biomarker for acromegaly using broad acoustic analysis and machine learning.

Methods: Voice recordings from patients with acromegaly and matched controls were collected using a mobile phone at Swedish university hospitals. Anthropometric and clinical data and the Voice Handicap Index (VHI) were assessed. Digital voice analysis of a sustained and stable vowel [a] resulted in 3274 parameters, which were used for training of machine learning models classifying the speaker as “acromegaly” or “control.” The machine learning models were trained with 76% of the data and the remaining 24% was used to assess their performance. For comparison, voice recordings of 50 pairs of participants were assessed by 12 experienced endocrinologists.

Results: We included 151 Swedish patients with acromegaly (13% biochemically active and 10% newly diagnosed) and 139 matched controls. The machine learning model identified patients with acromegaly more accurately (area under the receiver operating curve [ROC AUC] 0.84) than experienced endocrinologists (ROC AUC 0.69). Self-reported voice problems were more pronounced in patients with acromegaly than matched controls (median VHI 6 vs 2, P < .01) with higher prevalence of clinically significant voice handicap (VHI ≥20: 22.5% vs 3.6%).

Conclusion: Digital voice analysis can identify patients with acromegaly from short voice recordings with high accuracy. Patients with acromegaly experience more voice disorders than matched controls.

Background/aim: Acromegaly is a rare disease caused by a pituitary tumor secreting excess growth hormone, which leads to acral growth, organ enlargement, and facial changes. Patients with acromegaly have an increased risk of type 2 diabetes, cardiovascular disease, and arthropathy. Due to the rarity and slow progression of the disease, there is a considerable diagnostic delay (5–8 years), which contributes to increased morbidity and mortality. This thesis is based on four studies aimed at investigating the presentation of sleep apnea and carpal tunnel syndrome (CTS) in patients with acromegaly and the potential for digital analysis of voice and face to identify patients with acromegaly. 

Methods and results: Paper I was a cross-sectional, multicenter study of 259 patients with acromegaly: 29% of the patients were previously diagnosed with sleep apnea, with more than half (57%) of these diagnosed prior to the diagnosis of acromegaly. Another 8% of this cohort were found to have undiagnosed sleep apnea by targeted clinical assessment and sleep investigation. Paper II was a retrospective, national registry-based study of 556 patients with a diagnosis of acromegaly from the National Patient Registry. It found a 6-fold higher incidence of CTS diagnosis and surgery prior to acromegaly diagnosis compared to the general population. The risk of CTS was higher in women with acromegaly, and 84% of patients with CTS were diagnosed and surgically treated before the diagnosis of acromegaly. The potential window of opportunity to diagnose acromegaly earlier led us to investigate new non-invasive screening tools. In Paper III, a multicenter cohort study, we collected voice recordings from 151 patients with acromegaly (23% biochemically active) and 139 matched controls to create a machine learning algorithm, which identified the voice of patients with acromegaly at higher accuracy than experienced endocrinologists (ROC AUC 0.84 vs 0.69). Both biochemically active and controlled patients with acromegaly reported increased voice impairment (Voice Handicap Index) compared to controls. In Paper IV, we used facial images from 155 patients and 153 controls from the same cohort and machine learning algorithms for face analysis to train several machine learning models. The best model matched the accuracy of the compound assessment of 12 experienced endocrinologists (ROC AUC 0.85 vs 0.89) in acromegaly prediction while showing a higher sensitivity (0.82 vs 0.66). 

Conclusions: The diagnostic delay in acromegaly leads to the presence of comorbidities long before the disease is recognized. In this time window, non-invasive screening tools based on facial and voice analysis may improve the chances for earlier diagnosis. 

Purpose: The aim of this study was to establish a valid national cohort of patients diagnosed with acromegaly by combining data from the general National Patient Register (NPR) and the disease-specific Swedish Pituitary Register (SPR).

Methods: Patients ≥ 18 years of age at diagnosis of acromegaly reported from 1991 to 2018 who were registered in the NPR and/or SPR were included. The diagnosis of acromegaly was considered correct for patients identified in both registers or confirmed through chart review. Medical records were reviewed in two of Sweden´s six health care regions if the patient was reported only in the NPR. An algorithm for the NPR, with criteria requiring multiple diagnosis registrations and tumour and/or surgery codes, was constructed to reduce the number of patients to review in the remaining four regions.

Results: A total of 1866 patients were identified. Among these, 938 were reported in both registers. After application of the algorithm and chart review, the diagnosis was confirmed for 83 of the 906 patients found only in the NPR. Among 22 patients only registered in the SPR, a review of medical records confirmed acromegaly in 13. This resulted in a total of 1034 cases with acromegaly during the study period. The incidence rate of acromegaly in Sweden 1991–2018 was calculated to 4.0/million/year in the entire population and 5.1/million/year among subjects ≥ 18 years of age.

Conclusion: The combination of the SPR and NPR established a valid cohort of patients diagnosed with acromegaly and increased the estimated incidence in Sweden.

Objective: Carpal tunnel syndrome (CTS) is common in patients with acromegaly, with a reported prevalence of 19–64%.We studied CTS in a large national cohort of patients with acromegaly and the temporal relationship between thetwo diagnoses.

Design: Retrospective, nationwide, cohort study including patients diagnosed with acromegaly in Sweden, 2005–2017,identified in the Swedish Healthcare Registries.

Methods: CTS (diagnosis and surgery in specialised healthcare) was analysed from 8.5 years before the diagnosis ofacromegaly until death or end of the study. Standardised incidence ratios (SIRs) with 95% CIs were calculated for CTSwith the Swedish population as reference.

Results: The analysis included 556 patients with acromegaly (50% women) diagnosed at mean (s.d.) age 50.1 (15.0)years. During the study period, 48 patients were diagnosed with CTS and 41 patients underwent at least one CTSsurgery. In the latter group, 35 (85%) were operated for CTS before the acromegaly diagnosis; mean interval (range)2.2 (0.3–8.5) years and the SIR for having CTS surgery before the diagnosis of acromegaly was 6.6 (4.8–8.9). Womenwith acromegaly had a higher risk for CTS than men (hazard ratio: 2.5, 95% CI: 1.3–4.7).

Conclusions: Patients with acromegaly had a 6-fold higher incidence for CTS surgery before the diagnosis of acromegalycompared with the general population. The majority of patients with both diagnoses were diagnosed with CTS priorto acromegaly. Increased awareness of signs of acromegaly in patients with CTS might help to shorten the diagnosticdelay in acromegaly, especially in women.

Purpose:

Patients with acromegaly have an increased risk of sleep apnea, but reported prevalence rates vary largely. Here we aimed to evaluate the sleep apnea prevalence in a large national cohort of patients with acromegaly, to examine possible risk factors, and to assess the proportion of patients diagnosed with sleep apnea prior to acromegaly diagnosis.

Methods: Cross-sectional multicenter study of 259 Swedish patients with acromegaly. At patients' follow-up visits at the endocrine outpatient clinics of all seven university hospitals in Sweden, questionnaires were completed to assess previous sleep apnea diagnosis and treatment, cardiovascular diseases, smoking habits, anthropometric data, and S-IGF-1 levels. Daytime sleepiness was evaluated using the Epworth Sleepiness Scale. Patients suspected to have undiagnosed sleep apnea were referred for sleep apnea investigations.

Results: Of the 259 participants, 75 (29%) were diagnosed with sleep apnea before the study start. In 43 (57%) of these patients, sleep apnea had been diagnosed before the diagnosis of acromegaly. After clinical assessment and sleep studies, sleep apnea was diagnosed in an additional 20 patients, yielding a total sleep apnea prevalence of 37%. Higher sleep apnea risk was associated with higher BMI, waist circumference, and index finger circumference. Sleep apnea was more frequent among patients with S-IGF-1 levels in the highest quartile.

Conclusion: Sleep apnea is common among patients with acromegaly, and is often diagnosed prior to their acromegaly diagnosis. These results support early screening for sleep apnea in patients with acromegaly and awareness for acromegaly in patients with sleep apnea.

Background. There is a substantial delay in the diagnosis of acromegaly contributing to increased morbidity. Analysis of face images using machine learning to identify acromegaly has been investigated with impressive results but for clinical utilization there is a need for reassessment in different, large acromegaly cohorts using new methodological insights of machine learning development. 

Methods. Video recordings from patients with acromegaly and matched controls were collected at all Swedish University hospitals. Facial pictures from different angles were extracted. Clinical data regarding disease course and status were collected. Machine learning models based on four pre-trained deep neural networks (FaRL and 3 ImageNet models: ResNet50, InceptionV2 and DenseNet121) were trained to distinguish photographs of patients with acromegaly from controls. The models were compared to the performance of 12 experienced endocrinologists. 

Results. The FaRL based model matched the diagnostic precision of human experts (ROC AUC 0.89, balanced accuracy 0.89 for both) but had a higher sensitivity (0.82 vs 0.66) and comparable specificity (0.87 vs 0.93). An ensemble of the ImageNet models was inferior to FaRL (ROC AUC 0.85, sensitivity 0.75 and specificity 0.80). All models and human experts showed a higher sensitivity in the classification of males compared to females. There was a high level of agreement between FaRL, ImageNet ensemble, and the experts for true negatives (76%) but lower agreement for true positives (55%). 

Conclusions. The FaRL based model shows the same accuracy as expert endocrinologists in acromegaly identification by face photographs, but with a higher sensitivity. This supports that digital face analysis can be useful in acromegaly detection, but further studies are needed for validation and evaluation in a clinical setting.