Deep Learning Pitfall: Impact of Novel Ultrasound Equipment Introduction on Algorithm Performance and the Realities of Domain Adaptation

Inferior vena cava distensibility during pressure support ventilation: a prospective study evaluating interchangeability of subcostal and trans‑hepatic views, with both M‑mode and automatic border tracing

The Inferior Vena Cava (IVC) is commonly utilized to evaluate fluid status in the Intensive Care Unit (ICU),with more recent emphasis on the study of venous congestion. It is predominantly measured via subcostal approach (SC) or trans-hepatic (TH) views, and automated border tracking (ABT) software has been introduced to facilitate its assessment. Prospective observational study on patients ventilated in pressure support ventilation (PSV) with 2 × 2 factorial design. Primary outcome was to evaluate interchangeability of measurements of the IVC and the distensibility index (DI) obtained using both M-mode and ABT, across both SC and TH. Statistical analyses comprised Bland-Altman assessments for mean bias, limits of agreement (LoA), and the Spearman correlation coefficients. IVC visualization was 100% successful via SC, while TH view was unattainable in 17.4% of cases. As compared to the M-mode, the IVC-DI obtained through ABT approach showed divergences in both SC (mean bias 5.9%, LoA -18.4% to 30.2%, ICC = 0.52) and TH window (mean bias 6.2%, LoA -8.0% to 20.4%, ICC = 0.67). When comparing the IVC-DI measures obtained in the two anatomical sites, accuracy improved with a mean bias of 1.9% (M-mode) and 1.1% (ABT), but LoA remained wide (M-mode: -13.7% to 17.5%; AI: -19.6% to 21.9%). Correlation was generally suboptimal (r = 0.43 to 0.60). In PSV ventilated patients, we found that IVC-DI calculated with M-mode is not interchangeable with ABT measurements. Moreover, the IVC-DI gathered from SC or TH view produces not comparable results, mainly in terms of precision.

Self-Supervised Learning With Limited Labeled Data for Prostate Cancer Detection in High-Frequency Ultrasound

Deep learning-based analysis of high-frequency, high-resolution micro-ultrasound data shows great promise for prostate cancer (PCa) detection. Previous approaches to analysis of ultrasound data largely follow a supervised learning (SL) paradigm. Ground truth labels for ultrasound images used for training deep networks often include coarse annotations generated from the histopathological analysis of tissue samples obtained via biopsy. This creates inherent limitations on the availability and quality of labeled data, posing major challenges to the success of SL methods. However, unlabeled prostate ultrasound data are more abundant. In this work, we successfully apply self-supervised representation learning to micro-ultrasound data. Using ultrasound data from 1028 biopsy cores of 391 subjects obtained in two clinical centers, we demonstrate that feature representations learned with this method can be used to classify cancer from noncancer tissue, obtaining an AUROC score of 91% on an independent test set. To the best of our knowledge, this is the first successful end-to-end self-SL (SSL) approach for PCa detection using ultrasound data. Our method outperforms baseline SL approaches, generalizes well between different data centers, and scales well in performance as more unlabeled data are added, making it a promising approach for future research using large volumes of unlabeled data. Our code is publicly available at https://www.github.com/MahdiGilany/SSL_micro_ultrasound.

Inferior vena cava distensibility from subcostal and trans-hepatic imaging using both M-mode or artificial intelligence: a prospective study on mechanically ventilated patients

BACKGROUND

Variation of inferior vena cava (IVC) is used to predict fluid-responsiveness, but the IVC visualization with standard sagittal approach (SC, subcostal) cannot be always achieved. In such cases, coronal trans-hepatic (TH) window may offer an alternative, but the interchangeability of IVC measurements in SC and TH is not fully established. Furthermore, artificial intelligence (AI) with automated border detection may be of clinical value but it needs validation.

METHODS

Prospective observational validation study in mechanically ventilated patients with pressure-controlled mode. Primary outcome was the IVC distensibility (IVC-DI) in SC and TH imaging, with measurements taken both in M-Mode or with AI software. We calculated mean bias, limits of agreement (LoA), and intra-class correlation (ICC) coefficient.

RESULTS

Thirty-three patients were included. Feasibility rate was 87.9% and 81.8% for SC and TH visualization, respectively. Comparing imaging from the same anatomical site acquired with different modalities (M-Mode vs AI), we found the following IVC-DI differences: (1) SC: mean bias - 3.1%, LoA [- 20.1; 13.9], ICC = 0.65; (2) TH: mean bias - 2.0%, LoA [- 19.3; 15.4], ICC = 0.65. When comparing the results obtained from the same modality but from different sites (SC vs TH), IVC-DI differences were: (3) M-Mode: mean bias 1.1%, LoA [- 6.9; 9.1], ICC = 0.54; (4) AI: mean bias 2.0%, LoA [- 25.7; 29.7], ICC = 0.32.

CONCLUSIONS

In patients mechanically ventilated, AI software shows good accuracy (modest overestimation) and moderate correlation as compared to M-mode assessment of IVC-DI, both for SC and TH windows. However, precision seems suboptimal with wide LoA. The comparison of M-Mode or AI between different sites yields similar results but with weaker correlation. Trial registration Reference protocol: 53/2022/PO, approved on 21/03/2022.

Assessment of the inferior vena cava collapsibility from subcostal and trans-hepatic imaging using both M-mode or artificial intelligence: a prospective study on healthy volunteers

PURPOSE

Assessment of the inferior vena cava (IVC) respiratory variation may be clinically useful for the estimation of fluid-responsiveness and venous congestion; however, imaging from subcostal (SC, sagittal) region is not always feasible. It is unclear if coronal trans-hepatic (TH) IVC imaging provides interchangeable results. The use of artificial intelligence (AI) with automated border tracking may be helpful as part of point-of-care ultrasound but it needs validation.

METHODS

Prospective observational study conducted in spontaneously breathing healthy volunteers with assessment of IVC collapsibility (IVCc) in SC and TH imaging, with measures taken in M-mode or with AI software. We calculated mean bias and limits of agreement (LoA), and the intra-class correlation (ICC) coefficient with their 95% confidence intervals.

RESULTS

Sixty volunteers were included; IVC was not visualized in five of them (n = 2, both SC and TH windows, 3.3%; n = 3 in TH approach, 5%). Compared with M-mode, AI showed good accuracy both for SC (IVCc: bias - 0.7%, LoA [- 24.9; 23.6]) and TH approach (IVCc: bias 3.7%, LoA [- 14.9; 22.3]). The ICC coefficients showed moderate reliability: 0.57 [0.36; 0.73] in SC, and 0.72 [0.55; 0.83] in TH. Comparing anatomical sites (SC vs TH), results produced by M-mode were not interchangeable (IVCc: bias 13.9%, LoA [- 18.1; 45.8]). When this evaluation was performed with AI, such difference became smaller: IVCc bias 7.7%, LoA [- 19.2; 34.6]. The correlation between SC and TH assessments was poor for M-mode (ICC = 0.08 [- 0.18; 0.34]) while moderate for AI (ICC = 0.69 [0.52; 0.81]).

CONCLUSIONS

The use of AI shows good accuracy when compared with the traditional M-mode IVC assessment, both for SC and TH imaging. Although AI reduces differences between sagittal and coronal IVC measurements, results from these sites are not interchangeable.

Making Artificial Intelligence Lemonade Out of Data Lemons: Adaptation of a Public Apical Echo Database for Creation of a Subxiphoid Visual Estimation Automatic Ejection Fraction Machine Learning Algorithm

OBJECTIVES

A paucity of point-of-care ultrasound (POCUS) databases limits machine learning (ML). Assess feasibility of training ML algorithms to visually estimate left ventricular ejection fraction (EF) from a subxiphoid (SX) window using only apical 4-chamber (A4C) images.

METHODS

Researchers used a long-short-term-memory algorithm for image analysis. Using the Stanford EchoNet-Dynamic database of 10,036 A4C videos with calculated exact EF, researchers tested 3 ML training permeations. First, training on unaltered Stanford A4C videos, then unaltered and 90° clockwise (CW) rotated videos and finally unaltered, 90° rotated and horizontally flipped videos. As a real-world test, we obtained 615 SX videos from Harbor-UCLA (HUCLA) with EF calculations in 5% ranges. Researchers performed 1000 randomizations of EF point estimation within HUCLA EF ranges to compensate for ML and HUCLA EF mismatch, obtaining a mean value for absolute error (MAE) comparison and performed Bland-Altman analyses.

RESULTS

The ML algorithm EF mean MAE was estimated at 23.0, with a range of 22.8-23.3 using unaltered A4C video, mean MAE was 16.7, with a range of 16.5-16.9 using unaltered and 90° CW rotated video, mean MAE was 16.6, with a range of 16.3-16.8 using unaltered, 90° CW rotated and horizontally flipped video training. Bland-Altman showed weakest agreement at 40-45% EF.

CONCLUSIONS

Researchers successfully adapted unrelated ultrasound window data to train a POCUS ML algorithm with fair MAE using data manipulation to simulate a different ultrasound examination. This may be important for future POCUS algorithm design to help overcome a paucity of POCUS databases.

Machine learning algorithm using publicly available echo database for simplified “visual estimation” of left ventricular ejection fraction

BACKGROUND

Left ventricular ejection fraction calculation automation typically requires complex algorithms and is dependent of optimal visualization and tracing of endocardial borders. This significantly limits usability in bedside clinical applications, where ultrasound automation is needed most.

AIM

To create a simple deep learning (DL) regression-type algorithm to visually estimate left ventricular (LV) ejection fraction (EF) from a public database of actual patient echo examinations and compare results to echocardiography laboratory EF calculations.

METHODS

A simple DL architecture previously proven to perform well on ultrasound image analysis, VGG16, was utilized as a base architecture running within a long short term memory algorithm for sequential image (video) analysis. After obtaining permission to use the Stanford EchoNet-Dynamic database, researchers randomly removed approximately 15% of the approximately 10036 echo apical 4-chamber videos for later performance testing. All database echo examinations were read as part of comprehensive echocardiography study performance and were coupled with EF, end systolic and diastolic volumes, key frames and coordinates for LV endocardial tracing in csv file. To better reflect point-of-care ultrasound (POCUS) clinical settings and time pressure, the algorithm was trained on echo video correlated with calculated ejection fraction without incorporating additional volume, measurement and coordinate data. Seventy percent of the original data was used for algorithm training and 15% for validation during training. The previously randomly separated 15% (1263 echo videos) was used for algorithm performance testing after training completion. Given the inherent variability of echo EF measurement and field standards for evaluating algorithm accuracy, mean absolute error (MAE) and root mean square error (RMSE) calculations were made on algorithm EF results compared to Echo Lab calculated EF. Bland-Atlman calculation was also performed. MAE for skilled echocardiographers has been established to range from 4% to 5%.

RESULTS

The DL algorithm visually estimated EF had a MAE of 8.08% (95%CI 7.60 to 8.55) suggesting good performance compared to highly skill humans. The RMSE was 11.98 and correlation of 0.348.

CONCLUSION

This experimental simplified DL algorithm showed promise and proved reasonably accurate at visually estimating LV EF from short real time echo video clips. Less burdensome than complex DL approaches used for EF calculation, such an approach may be more optimal for POCUS settings once improved upon by future research and development.

Review of Machine Learning in Lung Ultrasound in COVID-19 Pandemic

Ultrasound imaging of the lung has played an important role in managing patients with COVID-19–associated pneumonia and acute respiratory distress syndrome (ARDS). During the COVID-19 pandemic, lung ultrasound (LUS) or point-of-care ultrasound (POCUS) has been a popular diagnostic tool due to its unique imaging capability and logistical advantages over chest X-ray and CT. Pneumonia/ARDS is associated with the sonographic appearances of pleural line irregularities and B-line artefacts, which are caused by interstitial thickening and inflammation, and increase in number with severity. Artificial intelligence (AI), particularly machine learning, is increasingly used as a critical tool that assists clinicians in LUS image reading and COVID-19 decision making. We conducted a systematic review from academic databases (PubMed and Google Scholar) and preprints on arXiv or TechRxiv of the state-of-the-art machine learning technologies for LUS images in COVID-19 diagnosis. Openly accessible LUS datasets are listed. Various machine learning architectures have been employed to evaluate LUS and showed high performance. This paper will summarize the current development of AI for COVID-19 management and the outlook for emerging trends of combining AI-based LUS with robotics, telehealth, and other techniques.