A single Chest X-ray Could Predict 10-Year CVD Risk

Researchers presented in the 2022 RSNA Annual meeting the results of a deep-learning model (AI) based on which a single chest x-ray could predict a patient's 10-year risk of dying from a heart attack or a stroke.

They tested the algorithm against a group of 11,430 outpatients, with an average age 60-years who underwent outpatient chest x-ray and were potentially eligible to receive statins.  Of those patients ,1096 or 9.6% had a major adverse cardiac event in a median 10-year follow up.

There was a significant association of the CVD risk found by chest x-ray and those found with MACE.

This is an example on how AI could detect clinically relevant outcomes with a widely used and low-cost screening test.

The paper was presented at 2022 RSNA meeting, Abstract T#-SSCH04-1

As today is the first of a new year I want to wish you all a happy and healthy 2023.  I would like to thank the 14,000+ who read the 174 posts I published the last 7 years.  

I also want to extend my sincere appreciation to three mentors of mine; my late Chairman Dr. V Capek MD and late Dean B. Siegel MD at UIC and Dean M Tzagournis MD at OSUMC for guiding and supporting me during my academic career. Also the many associates and trainees who daily enriched my experience in our specialty that had an explosive growth in the last 50 years.  

Finally, I want to let you know that starting next month the blog will have a new editor, a talented young radiologist, who with her ideas and knowledge of recent developments and advances in Radiology will make Radiology Monthly better for the benefit of all who read it.

Improved Fracture Recognition with AI assistance

 A retrospective study published in Radiology that included 489 patients with fractures that were interpreted by 24 readers showed a 10% improvement of fracture detection (75% vs 65%, superiority P<.001) when Artificial intelligence (AI) assistance was used.  AI also decreases the reading time by 6.3 seconds.  

The authors concluded that AI assistance improved the sensitivity and specificity of fracture detection for radiologists and non-radiologists alike  and shortened slightly their interpretation time. 

Single-view DBT and AI setup allows for effective screening mammography

A study published in Radiology determined that a single digital breast tomosynthesis (DBT) image in combination with artificial intelligence (AI) improves radiologists' productivity.

A retrospective study of 190 women with bilateral mediolateral oblique breast images that were acquired with a wide-angle DBT system was obtained.  The examination -based reader- average AUC was higher when interpreting results with AI support than when reading unaided.  The average sensitivity increased with AI support, whereas no differences in the specificity and reading time were detected.

The authors concluded that using a single-view DBT in conjunction with an AI setup could allow for more effective screening, especially in cancers detected, than using DBT alone.  

AI Matches Radiologists in Diagnosing Lung Cancer.

 A study published in Radiology found that deep learning artificial intelligence (AI) algorithm diagnosed lung cancer from chest radiographs at a rate similar to radiologists.  The researchers tested it on 10,285 radiographs form 10,202 individuals with 10 radiographs with visible cancer.  The algorithm showed comparable sensitivity 90% to 60% for the radiologists.  In a screening cohort of 100,525 chest x-rays from 50,070 individuals with 47 radiographs with lung cancer, the algorithm's sensitivity was 83% and false positive rate was 3%.  The investigators suggested the algorithm could prove useful especially for clinicians treating healthy persons with lower prevalence of lung cancer.

Artificial Intelligence (AI) Outperforms Radiologists in Mammography

The aim of screening mammography is to detect breast cancer in women as early as possible before signs of the disease become clinically obvious.  In a study published in Nature McKinney et al found that AI bested radiologists in detecting breast cancer in screening mammograms. 

Mammograms of 25,856 women in the United Kingdom and 3,097 women in the United States were used to train the AI system. AI was then used to identify the presence of breast cancer in mammograms of women who were known to have had either biopsy-proven breast cancer or normal follow-up imaging results at least 365 days later. The study included mammograms; by conventional digital (2D) mammography and tomosynthesis (also known as 3D mammography). 

The authors report that the AI system outperformed diagnoses made by the radiologists who initially interpreted the mammograms, and the decisions of 6 expert radiologists who interpreted 500 randomly selected cases.

The study reports an absolute reduction of 5.7% and 1.2% (USA and UK) in false positives and 9.4% and 2.7% in false negatives. In an independent study of six radiologists, the AI system outperformed all of the human readers: the area under the receiver operating characteristic curve (AUC-ROC) for the AI system was greater than the AUC-ROC for the average radiologist by an absolute margin of 11.5%. The authors also performed a simulation in which the AI system participated in the double-reading process that is common in the UK, and found that the AI system maintained non-inferior performance and reduced the workload of the second reader by 88%. 

The authors suggest that further assessment of the AI system with clinical trials may lead to improvements in the accuracy and efficiency of breast cancer screening by limiting the high rates of false positives and negatives which are known to take place in the interpretation of mammograms.

AI can diagnose Myocardial Infarction on non-enhanced MRI

A retrospective study of 212 patients published in Radiology used deep learning (AI) to identify and delineate chronic myocardial infarction without late gadolinium enhancement.

The model extracted motion features from the left ventricle on non-enhanced cardiac cine MRI and its per-segment sensitivity and specificity was 90% and 99 percent, therefore deep learning on non-enhanced cine cardiac MRI data can detect the presence and extent of chronic myocardial infarction. 

This approach has the potential to reduce the use of gadolinium contrast administration in patients with renal impairment, which is common in patients with coronary artery disease.

Artificial intelligence Can Predict Which Patients Will Develop Breast Cancer Within a Year

A study published in Radiology found that a deep learning artificial intelligence (AI) model from IBM Research can predict the development of malignant breast cancer in patients within a year by linking health records and mammograms. 

In this retrospective study, 52,936 images were obtained in 13,234 women who underwent at least one mammogram and who had health records for at least 1 year before undergoing mammography.  The algorithm was trained on 9,611 mammograms and health records to predict biopsy malignancy and to differentiate between normal from abnormal screenings.

The AI could correctly forecasted respective development of 87 percent and 77 percent of cancerous and non-cancerous cases, and also identified breast cancer in 48 percent of patients that otherwise would have been overlooked, with accuracy comparable to radiologists therefore it has the potential to substantially reduce missed diagnoses of breast cancer.

Artificial Intelligence is Useful in the Interpretation of Screening Mammograms

A study published in Radiology found that breast radiologists had a slight higher diagnostic performance when using artificial intelligence (AI) with no additional time required.

Screening digital mammograms from 240 women (median age, 62 years; range, 39–89 years) performed between 2013 and 2017 were analyzed in this study. The mammograms were interpreted with and without AI support.

The researchers found that the cancer detection improved for all breast densities, and was independent of lesion type, vendor image quality, when radiologists used AI and interestingly did not lengthen interpretation time. The radiologists’ detection slightly improved when using AI support, with the average area under the receiver operating characteristic curve (AUC) increasing form 0,87 to 0.89.  Sensitivity increased with AI support 86% vs. 83%, whereas specificity improved slightly 79% vs. 77%. Reading time per case was for all practical purposes identical  (unaided, 146 seconds; supported by AI, 149 seconds).

The researchers concluded that AI assisted interpreting radiologists and improved their cancer detection at mammography when using AI without adding to the interpetation time.

Artificial Intelligence Can Detect Wrist Fractures

A study published in Radiology: Artificial Intelligence found that convolutional neural networks could detect and show fractures on wrist radiographs with a high level of sensitivity and specificity.

A dataset of 7356 wrist radiographs was split into training (90%) and validation (10%) sets.  The models were tested on an unseen test set of 524 consecutive emergency wrist radiographic studies with two radiologists in consensus as the reference standard.

The model detected and correctly localized 310 (91.2%) of 340 and 236 (96.3%) of 245 of all radius and ulna fractures on the frontal and lateral views, respectively. The per-study sensitivity, specificity were 98.1%, and 72.9%, respectively.

The authors concluded that convolutional neural networks were able to detect and localize radius and ulna fractures on wrist radiographs with high sensitivity and specificity.