View this email in your browser

Doctor Penguin Weekly

Welcome to the fourth week for the Doctor Penguin newsletter! Over the past week, these AI papers caught our attention: 

Park et al. show that HeadXNet, a deep learning segmentation model, improves the performance of clinicians in identifying intracranial aneurysms from computed tomographic angiography examinations. Kather et al. show that deep residual learning can predict MSI, which determines whether patients with gastrointestinal cancer respond exceptionally well to immunotherapy, directly from H&E histology. Porter et al. use a neural network on recorded cough sounds and up to five-symptom input derived from patient/parent-reported history to detect common respiratory disorders in children. Jung et al. develop a machine learning model on survey data to identify Korean adolescents at risk of suicide.
-- Eric Topol & Pranav Rajpurkar  

Quick Links:

  1. Deep Learning–Assisted Diagnosis of Cerebral Aneurysms Using the HeadXNet Model
  2. Deep learning can predict microsatellite instability directly from histology in gastrointestinal cancer
  3. A prospective multicentre study testing the diagnostic accuracy of an automated cough sound centred analytic system for the identification of common respiratory disorders in children
  4. Prediction models for high risk of suicide in Korean adolescents using machine learning techniques
  5. Predicting cochlear dead regions in patients with hearing loss through a machine learning-based approach: A preliminary study


In JAMA network open

Importance : Deep learning has the potential to augment clinician performance in medical imaging interpretation and reduce time to diagnosis through automated segmentation. Few studies to date have explored this topic.

Objective : To develop and apply a neural network segmentation model (the HeadXNet model) capable of generating precise voxel-by-voxel predictions of intracranial aneurysms on head computed tomographic angiography (CTA) imaging to augment clinicians' intracranial aneurysm diagnostic performance.

Design, Setting, and Participants : In this diagnostic study, a 3-dimensional convolutional neural network architecture was developed using a training set of 611 head CTA examinations to generate aneurysm segmentations. Segmentation outputs from this support model on a test set of 115 examinations were provided to clinicians. Between August 13, 2018, and October 4, 2018, 8 clinicians diagnosed the presence of aneurysm on the test set, both with and without model augmentation, in a crossover design using randomized order and a 14-day washout period. Head and neck examinations performed between January 3, 2003, and May 31, 2017, at a single academic medical center were used to train, validate, and test the model. Examinations positive for aneurysm had at least 1 clinically significant, nonruptured intracranial aneurysm. Examinations with hemorrhage, ruptured aneurysm, posttraumatic or infectious pseudoaneurysm, arteriovenous malformation, surgical clips, coils, catheters, or other surgical hardware were excluded. All other CTA examinations were considered controls.

Main Outcomes and Measures : Sensitivity, specificity, accuracy, time, and interrater agreement were measured. Metrics for clinician performance with and without model augmentation were compared.

Results : The data set contained 818 examinations from 662 unique patients with 328 CTA examinations (40.1%) containing at least 1 intracranial aneurysm and 490 examinations (59.9%) without intracranial aneurysms. The 8 clinicians reading the test set ranged in experience from 2 to 12 years. Augmenting clinicians with artificial intelligence-produced segmentation predictions resulted in clinicians achieving statistically significant improvements in sensitivity, accuracy, and interrater agreement when compared with no augmentation. The clinicians' mean sensitivity increased by 0.059 (95% CI, 0.028-0.091; adjusted P = .01), mean accuracy increased by 0.038 (95% CI, 0.014-0.062; adjusted P = .02), and mean interrater agreement (Fleiss κ) increased by 0.060, from 0.799 to 0.859 (adjusted P = .05). There was no statistically significant change in mean specificity (0.016; 95% CI, -0.010 to 0.041; adjusted P = .16) and time to diagnosis (5.71 seconds; 95% CI, 7.22-18.63 seconds; adjusted P = .19).

Conclusions and Relevance : The deep learning model developed successfully detected clinically significant intracranial aneurysms on CTA. This suggests that integration of an artificial intelligence-assisted diagnostic model may augment clinician performance with dependable and accurate predictions and thereby optimize patient care.

Park Allison, Chute Chris, Rajpurkar Pranav, Lou Joe, Ball Robyn L, Shpanskaya Katie, Jabarkheel Rashad, Kim Lily H, McKenna Emily, Tseng Joe, Ni Jason, Wishah Fidaa, Wittber Fred, Hong David S, Wilson Thomas J, Halabi Safwan, Basu Sanjay, Patel Bhavik N, Lungren Matthew P, Ng Andrew Y, Yeom Kristen W


Deep learning can predict microsatellite instability directly from histology in gastrointestinal cancer.



In Nature medicine

Microsatellite instability determines whether patients with gastrointestinal cancer respond exceptionally well to immunotherapy. However, in clinical practice, not every patient is tested for MSI, because this requires additional genetic or immunohistochemical tests. Here we show that deep residual learning can predict MSI directly from H&E histology, which is ubiquitously available. This approach has the potential to provide immunotherapy to a much broader subset of patients with gastrointestinal cancer.

Kather Jakob Nikolas, Pearson Alexander T, Halama Niels, Jäger Dirk, Krause Jeremias, Loosen Sven H, Marx Alexander, Boor Peter, Tacke Frank, Neumann Ulf Peter, Grabsch Heike I, Yoshikawa Takaki, Brenner Hermann, Chang-Claude Jenny, Hoffmeister Michael, Trautwein Christian, Luedde Tom


A prospective multicentre study testing the diagnostic accuracy of an automated cough sound centred analytic system for the identification of common respiratory disorders in children.


In Respiratory research

BACKGROUND : The differential diagnosis of paediatric respiratory conditions is difficult and suboptimal. Existing diagnostic algorithms are associated with significant error rates, resulting in misdiagnoses, inappropriate use of antibiotics and unacceptable morbidity and mortality. Recent advances in acoustic engineering and artificial intelligence have shown promise in the identification of respiratory conditions based on sound analysis, reducing dependence on diagnostic support services and clinical expertise. We present the results of a diagnostic accuracy study for paediatric respiratory disease using an automated cough-sound analyser.

METHODS : We recorded cough sounds in typical clinical environments and the first five coughs were used in analyses. Analyses were performed using cough data and up to five-symptom input derived from patient/parent-reported history. Comparison was made between the automated cough analyser diagnoses and consensus clinical diagnoses reached by a panel of paediatricians after review of hospital charts and all available investigations.

RESULTS : A total of 585 subjects aged 29 days to 12 years were included for analysis. The Positive Percent and Negative Percent Agreement values between the automated analyser and the clinical reference were as follows: asthma (97, 91%); pneumonia (87, 85%); lower respiratory tract disease (83, 82%); croup (85, 82%); bronchiolitis (84, 81%).

CONCLUSION : The results indicate that this technology has a role as a high-level diagnostic aid in the assessment of common childhood respiratory disorders.

TRIAL REGISTRATION : Australian and New Zealand Clinical Trial Registry (retrospective) - ACTRN12618001521213 : 11.09.2018.

Porter Paul, Abeyratne Udantha, Swarnkar Vinayak, Tan Jamie, Ng Ti-Wan, Brisbane Joanna M, Speldewinde Deirdre, Choveaux Jennifer, Sharan Roneel, Kosasih Keegan, Della Phillip


Algorithm, Asthma, Bronchiolitis, Childhood, Cough, Croup, Diagnosis, Pneumonia, Respiratory

Prediction models for high risk of suicide in Korean adolescents using machine learning techniques.


In PloS one

OBJECTIVE : Suicide in adolescents is a major problem worldwide and previous history of suicide ideation and attempt represents the strongest predictors of future suicidal behavior. The aim of this study was to develop prediction model to identify Korean adolescents of high risk suicide (= who have history of suicide ideation/attempt in previous year) using machine learning techniques.

METHODS : A nationally representative dataset of Korea Youth Risk Behavior Web-based Survey (KYRBWS) was used (n = 59,984 of middle and high school students in 2017). The classification process was performed using machine learning techniques such as logistic regression (LR), random forest (RF), support vector machine (SVM), artificial neural network (ANN), and extreme gradient boosting (XGB).

RESULTS : A total of 7,443 adolescents (12.4%) had a previous history of suicidal ideation/attempt. In the multivariable analysis, sadness (odds ratio [OR], 6.41; 95% confidence interval [95% CI], 6.08-6.87), violence (OR, 2.32; 95% CI, 2.01-2.67), substance use (OR, 1.93; 95% CI, 1.52-2.45), and stress (OR, 1.63; 95% CI, 1.40-1.86) were associated factors. Taking into account 26 variables as predictors, the accuracy of models of machine learning techniques to predict the high-risk suicidal was comparable with that of LR; the accuracy was best in XGB (79.0%), followed by SVM (78.7%), LR (77.9%), RF (77.8%), and ANN (77.5%).

CONCLUSIONS : The machine leaning techniques showed comparable performance with LR to classify adolescents who have previous history of suicidal ideation/attempt. This model will hopefully serve as a foundation for decreasing future suicides as it enables early identification of adolescents at risk of suicide and modification of risk factors.

Jung Jun Su, Park Sung Jin, Kim Eun Young, Na Kyoung-Sae, Kim Young Jae, Kim Kwang Gi


Predicting cochlear dead regions in patients with hearing loss through a machine learning-based approach: A preliminary study.


In PloS one

We propose a machine learning (ML)-based model for predicting cochlear dead regions (DRs) in patients with hearing loss of various etiologies. Five hundred and fifty-five ears from 380 patients (3,770 test samples) diagnosed with sensorineural hearing loss (SNHL) were analyzed. A threshold-equalizing noise (TEN) test was applied to detect the presence of DRs. Data were collected on sex, age, side of the affected ear, hearing loss etiology, word recognition scores (WRS), and pure-tone thresholds at each frequency. According to the cause of hearing loss as diagnosed by the physician, we categorized the patients into six groups: 1) SNHL with unknown etiology; 2) sudden sensorineural hearing loss (SSNHL); 3) vestibular schwannoma (VS); 4) Meniere's disease (MD); 5) noise-induced hearing loss (NIHL); or 6) presbycusis or age-related hearing loss (ARHL). To develop a predictive model, we performed recursive partitioning and regression for classification, logistic regression, and random forest. The overall prevalence of one or more DRs in test ears was 20.36% (113 ears). Among the 3,770 test samples, the overall frequency-specific prevalence of DR was 6.7%. WRS, pure-tone thresholds at each frequency, disease type (VS or MD), and frequency information were useful for predicting DRs. Sex and age were not associated with detecting DRs. Based on these results, we suggest possible predictive factors for determining the presence of DRs. To improve the predictive power of the model, a more flexible model or more clinical features, such as the duration of hearing loss or risk factors for developing DRs, may be needed.

Chang Young-Soo, Park Heesung, Hong Sung Hwa, Chung Won-Ho, Cho Yang-Sun, Moon Il Joon


Follow Us on

This email was sent to <<Email Address *>>
why did I get this?    unsubscribe from this list    update subscription preferences
Stanford · 353 Serra Mall · Stanford, CA 94305-5008 · USA

Email Marketing Powered by Mailchimp