Receive a weekly summary and discussion of the top papers of the week by leading researchers in the field.

Radiology Radiology

Artificial Intelligence-Based Polyp Detection in Colonoscopy: Where Have We Been, Where Do We Stand, and Where Are We Headed?

In Visceral medicine

Background : In the past, image-based computer-assisted diagnosis and detection systems have been driven mainly from the field of radiology, and more specifically mammography. Nevertheless, with the availability of large image data collections (known as the "Big Data" phenomenon) in correlation with developments from the domain of artificial intelligence (AI) and particularly so-called deep convolutional neural networks, computer-assisted detection of adenomas and polyps in real-time during screening colonoscopy has become feasible.

Summary : With respect to these developments, the scope of this contribution is to provide a brief overview about the evolution of AI-based detection of adenomas and polyps during colonoscopy of the past 35 years, starting with the age of "handcrafted geometrical features" together with simple classification schemes, over the development and use of "texture-based features" and machine learning approaches, and ending with current developments in the field of deep learning using convolutional neural networks. In parallel, the need and necessity of large-scale clinical data will be discussed in order to develop such methods, up to commercially available AI products for automated detection of polyps (adenoma and benign neoplastic lesions). Finally, a short view into the future is made regarding further possibilities of AI methods within colonoscopy.

Key Messages : Research of image-based lesion detection in colonoscopy data has a 35-year-old history. Milestones such as the Paris nomenclature, texture features, big data, and deep learning were essential for the development and availability of commercial AI-based systems for polyp detection.

Wittenberg Thomas, Raithel Martin


AI, Adenoma and polyp detection, Artificial intelligence, Colonoscopy, History

Radiology Radiology

New advances in CT imaging of pancreas diseases: a narrative review.

In Gland surgery

Computed tomography (CT) plays a pivotal role as a diagnostic tool in many diagnostic and diffuse pancreatic diseases. One of the major limits of CT is related to the radiation exposure of young patients undergoing repeated examinations. Besides the standard CT protocol, the most recent technological advances, such as low-voltage acquisitions with high performance X-ray tubes and iterative reconstructions, allow for significant optimization of the protocol with dose reduction. The variety of CT tools are further expanded by the introduction of dual energy: the production of energy-selective images (i.e., virtual monochromatic images) improves the image contrast and lesion detection while the material-selective images (e.g., iodine maps or virtual unenhanced images) are valuable for lesion detection and dose reduction. The perfusion techniques provide diagnostic and prognostic information lesion and parenchymal vascularization and interstitium. Both dual energy and perfusion CT have the potential for pushing the limits of conventional CT from morphological evaluation to quantitative imaging applied to inflammatory and oncological diseases. Advances in post-processing of CT images, such as pancreatic volumetry, texture analysis and radiomics provide relevant information for pancreatic function but also for the diagnosis, management and prognosis of pancreatic neoplasms. Artificial intelligence is promising for optimization of the workflow in qualitative and quantitative analyses. Finally, basic concepts on the role of imaging on screening of pancreatic diseases will be provided.

Agostini Andrea, Borgheresi Alessandra, Bruno Federico, Natella Raffaele, Floridi Chiara, Carotti Marina, Giovagnoni Andrea


CT quantitative, Pancreas, dual energy CT, perfusion CT, texture analysis

General General

Updates in using a molecular classifier to identify usual interstitial pneumonia in conventional transbronchial lung biopsy samples.

In Breathe (Sheffield, England)

A molecular classifier using a machine-learning algorithm based on genomic data could provide an objective method to aid clinicians and multidisciplinary teams to establish the diagnosis of IPF in less-invasive transbronchial lung biopsy samples

Crespo Andrea, Alfaro Tiago, Somogyi Vivien, Kreuter Michael


Radiology Radiology

Prediction of Obstructive Lung Disease from Chest Radiographs via Deep Learning Trained on Pulmonary Function Data.

In International journal of chronic obstructive pulmonary disease ; h5-index 50.0

Background : Chronic obstructive pulmonary disease (COPD), the third leading cause of death worldwide, is often underdiagnosed.

Purpose : To develop machine learning methods to predict COPD using chest radiographs and a convolutional neural network (CNN) trained with near-concurrent pulmonary function test (PFT) data. Comparison is made to natural language processing (NLP) of the associated radiologist text reports.

Materials and Methods : This IRB-approved single-institution retrospective study uses 6749 two-view chest radiograph exams (2012-2017, 4436 unique subjects, 54% female, 46% male), same-day associated radiologist text reports, and PFT exams acquired within 180 days. The Image Model (Resnet18 pre-trained with ImageNet CNN) is trained using frontal and lateral radiographs and PFTs with 10% of the subjects for validation and 19% for testing. The NLP Model is trained using radiologist text reports and PFTs. The primary metric of model comparison is the area under the receiver operating characteristic curve (AUC).

Results : The Image Model achieves an AUC of 0.814 for prediction of obstructive lung disease (FEV1/FVC <0.7) from chest radiographs and performs better than the NLP Model (AUC 0.704, p<0.001) from radiologist text reports where FEV1 = forced expiratory volume in 1 second and FVC = forced vital capacity. The Image Model performs better for prediction of severe or very severe COPD (FEV1 <0.5) with an AUC of 0.837 versus the NLP model AUC of 0.770 (p<0.001).

Conclusion : A CNN Image Model trained on physiologic lung function data (PFTs) can be applied to chest radiographs for quantitative prediction of obstructive lung disease with good accuracy.

Schroeder Joyce D, Bigolin Lanfredi Ricardo, Li Tao, Chan Jessica, Vachet Clement, Paine Iii Robert, Srikumar Vivek, Tasdizen Tolga


chronic obstructive pulmonary disease, machine learning, natural language processing, quantitative image analysis

Public Health Public Health

Cervical cancer: Epidemiology, risk factors and screening.

In Chinese journal of cancer research = Chung-kuo yen cheng yen chiu

Cervical cancer is one of the leading causes of cancer death among females worldwide and its behavior epidemiologically likes a venereal disease of low infectiousness. Early age at first intercourse and multiple sexual partners have been shown to exert strong effects on risk. The wide differences in the incidence among different countries also influenced by the introduction of screening. Although the general picture remains one of decreasing incidence and mortality, there are signs of an increasing cervical cancer risk probably due to changes in sexual behavior. Smoking and human papillomavirus (HPV) 16/18 are currently important issues in a concept of multifactorial, stepwise carcinogenesis at the cervix uteri. Therefore, society-based preventive and control measures, screening activities and HPV vaccination are recommended. Cervical cancer screening methods have evolved from cell morphology observation to molecular testing. High-risk HPV genotyping and liquid-based cytology are common methods which have been widely recommended and used worldwide. In future, accurate, cheap, fast and easy-to-use methods would be more popular. Artificial intelligence also shows to be promising in cervical cancer screening by integrating image recognition with big data technology. Meanwhile, China has achieved numerous breakthroughs in cervical cancer prevention and control which could be a great demonstration for other developing and resource-limited areas. In conclusion, although cervical cancer threatens female health, it could be the first cancer that would be eliminated by human beings with comprehensive preventive and control strategy.

Zhang Shaokai, Xu Huifang, Zhang Luyao, Qiao Youlin


Cervical cancer, epidemiology, risk factors, screening

General General

Predicting adult neuroscience intensive care unit admission from emergency department triage using a retrospective, tabular-free text machine learning approach.

In Scientific reports ; h5-index 158.0

Early admission to the neurosciences intensive care unit (NSICU) is associated with improved patient outcomes. Natural language processing offers new possibilities for mining free text in electronic health record data. We sought to develop a machine learning model using both tabular and free text data to identify patients requiring NSICU admission shortly after arrival to the emergency department (ED). We conducted a single-center, retrospective cohort study of adult patients at the Mount Sinai Hospital, an academic medical center in New York City. All patients presenting to our institutional ED between January 2014 and December 2018 were included. Structured (tabular) demographic, clinical, bed movement record data, and free text data from triage notes were extracted from our institutional data warehouse. A machine learning model was trained to predict likelihood of NSICU admission at 30 min from arrival to the ED. We identified 412,858 patients presenting to the ED over the study period, of whom 1900 (0.5%) were admitted to the NSICU. The daily median number of ED presentations was 231 (IQR 200-256) and the median time from ED presentation to the decision for NSICU admission was 169 min (IQR 80-324). A model trained only with text data had an area under the receiver-operating curve (AUC) of 0.90 (95% confidence interval (CI) 0.87-0.91). A structured data-only model had an AUC of 0.92 (95% CI 0.91-0.94). A combined model trained on structured and text data had an AUC of 0.93 (95% CI 0.92-0.95). At a false positive rate of 1:100 (99% specificity), the combined model was 58% sensitive for identifying NSICU admission. A machine learning model using structured and free text data can predict NSICU admission soon after ED arrival. This may potentially improve ED and NSICU resource allocation. Further studies should validate our findings.

Klang Eyal, Kummer Benjamin R, Dangayach Neha S, Zhong Amy, Kia M Arash, Timsina Prem, Cossentino Ian, Costa Anthony B, Levin Matthew A, Oermann Eric K