Accurate assessment of the lung sliding artefact on lung ultrasonography using a deep learning approach

Improving the Generalizability and Performance of an Ultrasound Deep Learning Model Using Limited Multicenter Data for Lung Sliding Artifact Identification

Deep learning (DL) models for medical image classification frequently struggle to generalize to data from outside institutions. Additional clinical data are also rarely collected to comprehensively assess and understand model performance amongst subgroups. Following the development of a single-center model to identify the lung sliding artifact on lung ultrasound (LUS), we pursued a validation strategy using external LUS data. As annotated LUS data are relatively scarce-compared to other medical imaging data-we adopted a novel technique to optimize the use of limited external data to improve model generalizability. Externally acquired LUS data from three tertiary care centers, totaling 641 clips from 238 patients, were used to assess the baseline generalizability of our lung sliding model. We then employed our novel Threshold-Aware Accumulative Fine-Tuning (TAAFT) method to fine-tune the baseline model and determine the minimum amount of data required to achieve predefined performance goals. A subgroup analysis was also performed and Grad-CAM++ explanations were examined. The final model was fine-tuned on one-third of the external dataset to achieve 0.917 sensitivity, 0.817 specificity, and 0.920 area under the receiver operator characteristic curve (AUC) on the external validation dataset, exceeding our predefined performance goals. Subgroup analyses identified LUS characteristics that most greatly challenged the model's performance. Grad-CAM++ saliency maps highlighted clinically relevant regions on M-mode images. We report a multicenter study that exploits limited available external data to improve the generalizability and performance of our lung sliding model while identifying poorly performing subgroups to inform future iterative improvements. This approach may contribute to efficiencies for DL researchers working with smaller quantities of external validation data.

Artificial Intelligence-Based Left Ventricular Ejection Fraction by Medical Students for Mortality and Readmission Prediction

INTRODUCTION

Point-of-care ultrasound has become a universal practice, employed by physicians across various disciplines, contributing to diagnostic processes and decision-making.

AIM

To assess the association of reduced (<50%) left-ventricular ejection fraction (LVEF) based on prospective point-of-care ultrasound operated by medical students using an artificial intelligence (AI) tool and 1-year primary composite outcome, including mortality and readmission for cardiovascular-related causes.

METHODS

Eight trained medical students used a hand-held ultrasound device (HUD) equipped with an AI-based tool for automatic evaluation of the LVEF of non-selected patients hospitalized in a cardiology department from March 2019 through March 2020.

RESULTS

The study included 82 patients (72 males aged 58.5 ± 16.8 years), of whom 34 (41.5%) were diagnosed with AI-based reduced LVEF. The rates of the composite outcome were higher among patients with reduced systolic function compared to those with preserved LVEF (41.2% vs. 16.7%, p = 0.014). Adjusting for pertinent variables, reduced LVEF independently predicted the composite outcome (HR 2.717, 95% CI 1.083-6.817, p = 0.033). As compared to those with LVEF ≥ 50%, patients with reduced LVEF had a longer length of stay and higher rates of the secondary composite outcome, including in-hospital death, advanced ventilatory support, shock, and acute decompensated heart failure.

CONCLUSION

AI-based assessment of reduced systolic function in the hands of medical students, independently predicted 1-year mortality and cardiovascular-related readmission and was associated with unfavorable in-hospital outcomes. AI utilization by novice users may be an important tool for risk stratification for hospitalized patients.

Lung ultrasound training: how short is too short? observational study on the effects of a focused theoretical training for novice learners

BACKGROUND

Lung ultrasound has been increasingly used in the last years for the assessment of patients with respiratory diseases; it is considered a simple technique, now spreading from physicians to other healthcare professionals as nurses and physiotherapists, as well as to medical students. These providers may require a different training to acquire lung ultrasound skills, since they are expected to have no previous experience with ultrasound. The aim of the study was to assess the impact of a short theoretical training focused on lung ultrasound pattern recognition in a population of novice nurse learners with no previous experience with ultrasound.

METHODS

We included the nurses attending a critical care advanced course for nurses performed at the University of Pavia. Images' interpretation skills were tested on two slide sets (a 25-clip set focused on B-pattern recognition and a 25-clip set focused on identification of pleural movement as lung sliding, lung pulse, lung point, no movement) before and after three 30-minute teaching modules dedicated to general ultrasound principles, B-lines assessment and lung sliding assessment. A cut off of 80% was considered acceptable for correctly interpreted images after this basic course.

RESULTS

22 nurses were enrolled (age 26.0 [24.0-28.0] years; men 4 (18%)); one nurse had previous experience with other ultrasound techniques, none of them had previous experience with lung ultrasound. After the training, the number of correctly interpreted clips improved from 3.5 [0.0-13.0] to 22.0 [19.0-23.0] (p < 0.0001) for B-pattern and from 0.5 [0.0-2.0] to 8.5 [6.0-12.0] (p < 0.0001) for lung sliding assessment. The number of correct answers for B-pattern recognition was significantly higher than for lung sliding assessment, both before (3.5 [0.0-13.0] vs. 0.5 [0.0-2.0]; p = 0.0036) and after (22.0 [19.0-23.0] vs. 8.5 [6.0-12.0]; p < 0.0001) the training. After the training, nurses were able to correctly recognize the presence or the absence of a B-pattern in 84.2 ± 10.3% of cases; lung sliding was correctly assessed in 37.1 ± 15.3% of cases.

CONCLUSIONS

Lung ultrasound is considered a simple technique; while a short, focused training significantly improves B-pattern recognition, lung sliding assessment may require a longer training for novice learners.

TRIAL REGISTRATION

Not applicable.

A scoping review of interpretability and explainability concerning artificial intelligence methods in medical imaging

PURPOSE

To review eXplainable Artificial Intelligence/(XAI) methods available for medical imaging/(MI).

METHOD

A scoping review was conducted following the Joanna Briggs Institute's methodology. The search was performed on Pubmed, Embase, Cinhal, Web of Science, BioRxiv, MedRxiv, and Google Scholar. Studies published in French and English after 2017 were included. Keyword combinations and descriptors related to explainability, and MI modalities were employed. Two independent reviewers screened abstracts, titles and full text, resolving differences through discussion.

RESULTS

228 studies met the criteria. XAI publications are increasing, targeting MRI (n = 73), radiography (n = 47), CT (n = 46). Lung (n = 82) and brain (n = 74) pathologies, Covid-19 (n = 48), Alzheimer's disease (n = 25), brain tumors (n = 15) are the main pathologies explained. Explanations are presented visually (n = 186), numerically (n = 67), rule-based (n = 11), textually (n = 11), and example-based (n = 6). Commonly explained tasks include classification (n = 89), prediction (n = 47), diagnosis (n = 39), detection (n = 29), segmentation (n = 13), and image quality improvement (n = 6). The most frequently provided explanations were local (78.1 %), 5.7 % were global, and 16.2 % combined both local and global approaches. Post-hoc approaches were predominantly employed. The used terminology varied, sometimes indistinctively using explainable (n = 207), interpretable (n = 187), understandable (n = 112), transparent (n = 61), reliable (n = 31), and intelligible (n = 3).

CONCLUSION

The number of XAI publications in medical imaging is increasing, primarily focusing on applying XAI techniques to MRI, CT, and radiography for classifying and predicting lung and brain pathologies. Visual and numerical output formats are predominantly used. Terminology standardisation remains a challenge, as terms like "explainable" and "interpretable" are sometimes being used indistinctively. Future XAI development should consider user needs and perspectives.

Point-of-care AI-assisted stepwise ultrasound pneumothorax diagnosis

Objective. Ultrasound is extensively utilized as a convenient and cost-effective method in emergency situations. Unfortunately, the limited availability of skilled clinicians in emergency hinders the wider adoption of point-of-care ultrasound. To overcome this challenge, this paper aims to aid less experienced healthcare providers in emergency lung ultrasound scans.Approach. To assist healthcare providers, it is important to have a comprehensive model that can automatically guide the entire process of lung ultrasound based on the clinician's workflow. In this paper, we propose a framework for diagnosing pneumothorax using artificial intelligence (AI) assistance. Specifically, the proposed framework for lung ultrasound scan follows the steps taken by skilled physicians. It begins with finding the appropriate transducer position on the chest to locate the pleural line accurately in B-mode. The next step involves acquiring temporal M-mode data to determine the presence of lung sliding, a crucial indicator for pneumothorax. To mimic the sequential process of clinicians, two DL models were developed. The first model focuses on quality assurance (QA) and regression of the pleural line region-of-interest, while the second model classifies lung sliding. To achieve the inference on a mobile device, a size of EfficientNet-Lite0 model was further reduced to have fewer than 3 million parameters.Main results. The results showed that both the QA and lung sliding classification models achieved over 95% in area under the receiver operating characteristic (AUC), while the ROI performance reached 89% in the dice similarity coefficient. The entire stepwise pipeline was simulated using retrospective data, yielding an AUC of 89%.Significance. The step-wise AI framework for the pneumothorax diagnosis with QA offers an intelligible guide for each clinical workflow, which achieved significantly high precision and real-time inferences.

Lung Ultrasound Reduces Chest X-rays in Postoperative Care after Thoracic Surgery: Is There a Role for Artificial Intelligence?-Systematic Review

BACKGROUND

Chest X-ray (CXR) remains the standard imaging modality in postoperative care after non-cardiac thoracic surgery. Lung ultrasound (LUS) showed promising results in CXR reduction. The aim of this review was to identify areas where the evaluation of LUS videos by artificial intelligence could improve the implementation of LUS in thoracic surgery.

METHODS

A literature review of the replacement of the CXR by LUS after thoracic surgery and the evaluation of LUS videos by artificial intelligence after thoracic surgery was conducted in Medline.

RESULTS

Here, eight out of 10 reviewed studies evaluating LUS in CXR reduction showed that LUS can reduce CXR without a negative impact on patient outcome after thoracic surgery. No studies on the evaluation of LUS signs by artificial intelligence after thoracic surgery were found.

CONCLUSION

LUS can reduce CXR after thoracic surgery. We presume that artificial intelligence could help increase the LUS accuracy, objectify the LUS findings, shorten the learning curve, and decrease the number of inconclusive results. To confirm this assumption, clinical trials are necessary. This research is funded by the Slovak Research and Development Agency, grant number APVV 20-0232.