The artificial intelligence evidence-based medicine pyramid

The Future of Breast Cancer Organized Screening Program Through Artificial Intelligence: A Scoping Review

Objective: The aim of this scoping review was to evaluate whether artificial intelligence integrated into breast cancer screening work strategies could help resolve some diagnostic issues that still remain. Methods: PubMed, Web of Science, and Scopus were consulted. The literature research was updated to 28 May 2024. The PRISMA method of selecting articles was used. The articles were classified according to the type of publication (meta-analysis, trial, prospective, and retrospective studies); moreover, retrospective studies were based on citizen recruitment (organized screening vs. spontaneous screening and a combination of both). Results: Meta-analyses showed that AI had an effective reduction in the radiologists' reading time of radiological images, with a variation from 17 to 91%. Furthermore, they highlighted how the use of artificial intelligence software improved the diagnostic accuracy. Systematic review speculated that AI could reduce false negatives and positives and detect subtle abnormalities missed by human observers. DR with AI results from organized screening showed a higher recall rate, specificity, and PPV. Data from opportunistic screening found that AI could reduce interval cancer with a corresponding reduction in serious outcome. Nevertheless, the analysis of this review suggests that the study of breast density and interval cancer still requires numerous applications. Conclusions: Artificial intelligence appears to be a promising technology for health, with consequences that can have a major impact on healthcare systems. Where screening is opportunistic and involves only one human reader, the use of AI can increase diagnostic performance enough to equal that of double human reading.

The Use of Wearable Sensors and Machine Learning Methods to Estimate Biomechanical Characteristics During Standing Posture or Locomotion: A Systematic Review

Balance deficits are present in a variety of clinical populations and can negatively impact quality of life. The integration of wearable sensors and machine learning technology (ML) provides unique opportunities to quantify biomechanical characteristics related to balance outside of a laboratory setting. This article provides a general overview of recent developments in using wearable sensors and ML to estimate or predict biomechanical characteristics such as center of pressure (CoP) and center of mass (CoM) motion. This systematic review was conducted according to PRISMA guidelines. Databases including Scopus, PubMed, CINHAL, Trip PRO, Cochrane, and Otseeker databases were searched for publications on the use of wearable sensors combined with ML to predict biomechanical characteristics. Fourteen publications met the inclusion criteria and were included in this review. From each publication, information on study characteristics, testing conditions, ML models applied, estimated biomechanical characteristics, and sensor positions were extracted. Additionally, the study type, level of evidence, and Downs and Black scale score were reported to evaluate methodological quality and bias. Most studies tested subjects during walking and utilized some type of neural network (NN) ML model to estimate biomechanical characteristics. Many of the studies focused on minimizing the necessary number of sensors and placed them on areas near or below the waist. Nearly all studies reporting RMSE and correlation coefficients had values <15% and >0.85, respectively, indicating strong ML model estimation accuracy. Overall, this review can help guide the future development of ML algorithms and wearable sensor technologies to estimate postural mechanics.