scispace - formally typeset
Search or ask a question

Showing papers by "Akshay S. Chaudhari published in 2021"


Journal ArticleDOI
TL;DR: A framework for important considerations that consist of appropriate model training paradigms, evaluation of model robustness, downstream clinical utility, opportunities for future advances, as well recommendations for best current practices is provided.
Abstract: Artificial intelligence algorithms based on principles of deep learning (DL) have made a large impact on the acquisition, reconstruction, and interpretation of MRI data. Despite the large number of retrospective studies using DL, there are fewer applications of DL in the clinic on a routine basis. To address this large translational gap, we review the recent publications to determine three major use cases that DL can have in MRI, namely, that of model-free image synthesis, model-based image reconstruction, and image or pixel-level classification. For each of these three areas, we provide a framework for important considerations that consist of appropriate model training paradigms, evaluation of model robustness, downstream clinical utility, opportunities for future advances, as well recommendations for best current practices. We draw inspiration for this framework from advances in computer vision in natural imaging as well as additional healthcare fields. We further emphasize the need for reproducibility of research studies through the sharing of datasets and software. LEVEL OF EVIDENCE: 5 TECHNICAL EFFICACY STAGE: 2.

47 citations


Journal ArticleDOI
10 Feb 2021
TL;DR: Diverse networks learned to segment the knee similarly, where high segmentation accuracy did not correlate with cartilage thickness accuracy and voting ensembles did not exceed individual network performance.
Abstract: A multi-institute challenge for knee MRI segmentation was organized, in which a generalized framework for characterizing and evaluating the semantic and clinical efficacy of automatic segmentation

45 citations


Journal ArticleDOI
TL;DR: qDESS prospectively enhanced with deep learning had strong inter-reader agreement with conventional knee MRI and near-equivalent diagnostic performance with respect to arthroscopy, and the ability of qDESS to automatically generate T2 maps increases sensitivity for cartilage abnormalities.
Abstract: Please see the Editorial Comment by Derik L. Davis discussing this article. BACKGROUND. Potential approaches for abbreviated knee MRI, including prospective acceleration with deep learning, have ac...

28 citations


Journal ArticleDOI
23 Aug 2021
TL;DR: In this paper, a deep learning-based image-quality enhancement algorithm was applied to fourfold reduced-count whole-body PET in a realistic clinical oncologic imaging environment with multiple blinded readers, institutions, and scanner types.
Abstract: More widespread use of positron emission tomography (PET) imaging is limited by its high cost and radiation dose. Reductions in PET scan time or radiotracer dosage typically degrade diagnostic image quality (DIQ). Deep-learning-based reconstruction may improve DIQ, but such methods have not been clinically evaluated in a realistic multicenter, multivendor environment. In this study, we evaluated the performance and generalizability of a deep-learning-based image-quality enhancement algorithm applied to fourfold reduced-count whole-body PET in a realistic clinical oncologic imaging environment with multiple blinded readers, institutions, and scanner types. We demonstrate that the low-count-enhanced scans were noninferior to the standard scans in DIQ (p < 0.05) and overall diagnostic confidence (p < 0.001) independent of the underlying PET scanner used. Lesion detection for the low-count-enhanced scans had a high patient-level sensitivity of 0.94 (0.83–0.99) and specificity of 0.98 (0.95–0.99). Interscan kappa agreement of 0.85 was comparable to intrareader (0.88) and pairwise inter-reader agreements (maximum of 0.72). SUV quantification was comparable in the reference regions and lesions (lowest p-value=0.59) and had high correlation (lowest CCC = 0.94). Thus, we demonstrated that deep learning can be used to restore diagnostic image quality and maintain SUV accuracy for fourfold reduced-count PET scans, with interscan variations in lesion depiction, lower than intra- and interreader variations. This method generalized to an external validation set of clinical patients from multiple institutions and scanner types. Overall, this method may enable either dose or exam-duration reduction, increasing safety and lowering the cost of PET imaging.

26 citations


Journal ArticleDOI
TL;DR: The whole-brain mean absolute discrepancy in cortical surface positioning and thickness estimation is below 100 μm at the single-subject level and below 50 μM at the group level for the simulated data, making the accuracy of cortical surfaces derived from super-resolution sufficient for most applications.
Abstract: Accurate and automated reconstruction of the in vivo human cerebral cortical surface from anatomical magnetic resonance (MR) images facilitates the quantitative analysis of cortical structure. Anatomical MR images with sub-millimeter isotropic spatial resolution improve the accuracy of cortical surface and thickness estimation compared to the standard 1-millimeter isotropic resolution. Nonetheless, sub-millimeter resolution acquisitions require averaging multiple repetitions to achieve sufficient signal-to-noise ratio and are therefore long and potentially vulnerable to subject motion. We address this challenge by synthesizing sub-millimeter resolution images from standard 1-millimeter isotropic resolution images using a data-driven supervised machine learning-based super-resolution approach achieved via a deep convolutional neural network. We systematically characterize our approach using a large-scale simulated dataset and demonstrate its efficacy in empirical data. The super-resolution data provide improved cortical surfaces similar to those obtained from native sub-millimeter resolution data. The whole-brain mean absolute discrepancy in cortical surface positioning and thickness estimation is below 100 μm at the single-subject level and below 50 μm at the group level for the simulated data, and below 200 μm at the single-subject level and below 100 μm at the group level for the empirical data, making the accuracy of cortical surfaces derived from super-resolution sufficient for most applications.

21 citations


Journal ArticleDOI
TL;DR: The 2D U-Net-based automated segmentation method yielded high agreement compared with manual segmentation and also demonstrated high accuracy and longitudinal test–retest reproducibility for morphometric analysis of articular cartilage derived from it, using both corFLASH and sagDESS.
Abstract: To evaluate the agreement, accuracy, and longitudinal reproducibility of quantitative cartilage morphometry from 2D U-Net-based automated segmentations for 3T coronal fast low angle shot (corFLASH) and sagittal double echo at steady-state (sagDESS) MRI. 2D U-Nets were trained using manual, quality-controlled femorotibial cartilage segmentations available for 92 Osteoarthritis Initiative healthy reference cohort participants from both corFLASH and sagDESS (n = 50/21/21 training/validation/test-set). Cartilage morphometry was computed from automated and manual segmentations for knees from the test-set. Agreement and accuracy were evaluated from baseline visits (dice similarity coefficient: DSC, correlation analysis, systematic offset). The longitudinal reproducibility was assessed from year-1 and -2 follow-up visits (root-mean-squared coefficient of variation, RMSCV%). Automated segmentations showed high agreement (DSC 0.89–0.92) and high correlations (r ≥ 0.92) with manual ground truth for both corFLASH and sagDESS and only small systematic offsets (≤ 10.1%). The automated measurements showed a similar test–retest reproducibility over 1 year (RMSCV% 1.0–4.5%) as manual measurements (RMSCV% 0.5–2.5%). The 2D U-Net-based automated segmentation method yielded high agreement compared with manual segmentation and also demonstrated high accuracy and longitudinal test–retest reproducibility for morphometric analysis of articular cartilage derived from it, using both corFLASH and sagDESS.

19 citations


Proceedings Article
18 Jul 2021
TL;DR: In this paper, the robustness of deep learning-based image reconstruction methods is investigated. And the authors find that both trained and un-trained methods are vulnerable to adversarial perturbations.
Abstract: Deep neural networks give state-of-the-art accuracy for reconstructing images from few and noisy measurements, a problem arising for example in accelerated magnetic resonance imaging (MRI). However, recent works have raised concerns that deep-learning-based image reconstruction methods are sensitive to perturbations and are less robust than traditional methods: Neural networks (i) may be sensitive to small, yet adversarially-selected perturbations, (ii) may perform poorly under distribution shifts, and (iii) may fail to recover small but important features in an image. In order to understand the sensitivity to such perturbations, in this work, we measure the robustness of different approaches for image reconstruction including trained and un-trained neural networks as well as traditional sparsity-based methods. We find, contrary to prior works, that both trained and un-trained methods are vulnerable to adversarial perturbations. Moreover, both trained and un-trained methods tuned for a particular dataset suffer very similarly from distribution shifts. Finally, we demonstrate that an image reconstruction method that achieves higher reconstruction quality, also performs better in terms of accurately recovering fine details. Our results indicate that the state-of-the-art deep-learning-based image reconstruction methods provide improved performance than traditional methods without compromising robustness.

9 citations


Posted Content
TL;DR: TorchXRayVision as discussed by the authors is an open source software library for working with chest X-ray datasets and deep learning models, which provides a common interface and common pre-processing chain for a wide set of publicly available chest X -ray datasets.
Abstract: TorchXRayVision is an open source software library for working with chest X-ray datasets and deep learning models. It provides a common interface and common pre-processing chain for a wide set of publicly available chest X-ray datasets. In addition, a number of classification and representation learning models with different architectures, trained on different data combinations, are available through the library to serve as baselines or feature extractors.

9 citations


Journal ArticleDOI
TL;DR: In this article, the authors present possible solutions for achieving clinically acceptable performance of image quantitation methods and briefly discuss the impact of minimizing variance and achieving generalizability towards clinical implementation and adoption.
Abstract: Image quantitation methods including quantitative MRI, multiparametric MRI, and radiomics offer great promise for clinical use. However, many of these methods have limited clinical adoption, in part due to issues of generalizability, that is, the ability to translate methods and models across institutions. Researchers can assess generalizability through measurement of repeatability and reproducibility, thus quantifying different aspects of measurement variance. In this article, we review the challenges to ensuring repeatability and reproducibility of image quantitation methods as well as present strategies to minimize their variance to enable wider clinical implementation. We present possible solutions for achieving clinically acceptable performance of image quantitation methods and briefly discuss the impact of minimizing variance and achieving generalizability towards clinical implementation and adoption.

7 citations



Journal ArticleDOI
26 May 2021
TL;DR: With use of a neural network-based cGAN approach, it is feasible to synthesize T2 maps in femoral cartilage from anatomic MRI sequences, giving good agreement with MESE scans.
Abstract: A neural network was developed to produce synthetic T2 maps of cartilage based on anatomic MRI scans that were not designed for T2 mapping.

Journal ArticleDOI
TL;DR: Low diffusion-weighted qDESS (qDESSLow) shows good agreement to contrast-enhanced MRI for characterization of the severity of synovitis, with a relative bias towards underestimation of severity.
Abstract: To determine whether synovitis graded by radiologists using hybrid quantitative double-echo in steady-state (qDESS) images can be utilized as a non-contrast approach to assess synovitis in the knee, compared against the reference standard of contrast-enhanced MRI (CE-MRI). Twenty-two knees (11 subjects) with moderate to severe osteoarthritis (OA) were scanned using CE-MRI, qDESS with a high diffusion weighting (qDESSHigh), and qDESS with a low diffusion weighting (qDESSLow). Four radiologists graded the overall impression of synovitis, their diagnostic confidence, and regional grading of synovitis severity at four sites (suprapatellar pouch, intercondylar notch, and medial and lateral peripatellar recesses) in the knee using a 4-point scale. Agreement between CE-MRI and qDESS, inter-rater agreement, and intra-rater agreement were assessed using a linearly weighted Gwet’s AC2. Good agreement was seen between CE-MRI and both qDESSLow (AC2 = 0.74) and qDESSHigh (AC2 = 0.66) for the overall impression of synovitis, but both qDESS sequences tended to underestimate the severity of synovitis compared to CE-MRI. Good inter-rater agreement was seen for both qDESS sequences (AC2 = 0.74 for qDESSLow, AC2 = 0.64 for qDESSHigh), and good intra-rater agreement was seen for both sequences as well (qDESSLow AC2 = 0.78, qDESSHigh AC2 = 0.80). Diagnostic confidence was moderate to high for qDESSLow (mean = 2.36) and slightly less than moderate for qDESSHigh (mean = 1.86), compared to mostly high confidence for CE-MRI (mean = 2.73). qDESS shows potential as an alternative MRI technique for assessing the severity of synovitis without the use of a gadolinium-based contrast agent.

Journal ArticleDOI
TL;DR: Sarcopenia is defined as the loss of muscle mass, strength, and function in patients with rheumatic disorders as mentioned in this paper, and it can be diagnosed using CT, MRI, and ultrasound.
Abstract: Sarcopenia is defined as the loss of muscle mass, strength, and function. Increasing evidence shows that sarcopenia is common in patients with rheumatic disorders. Although sarcopenia can be diagnosed using bioelectrical impedance analysis or DXA, increasingly it is diagnosed using CT, MRI, and ultrasound. In rheumatic patients, CT and MRI allow "opportunistic" measurement of body composition, including surrogate markers of sarcopenia, from studies obtained during routine patient care. Recognition of sarcopenia is important in rheumatic patients because sarcopenia can be associated with disease progression and poor outcomes. This article reviews how opportunistic evaluation of sarcopenia in rheumatic patients can be accomplished and potentially contribute to improved patient care.

Journal ArticleDOI
06 Oct 2021
TL;DR: Deep learning can enable a reduction in dose of fluorine 18 fluorodeoxyglucose in integrated PET/MRI scans of children and young adults with lymphoma for treatment response assessment without compr... as discussed by the authors.
Abstract: Deep learning may enable a reduction in dose of fluorine 18 fluorodeoxyglucose in integrated PET/MRI scans of children and young adults with lymphoma for treatment response assessment without compr...

Posted Content
TL;DR: In this paper, the robustness of deep learning-based image reconstruction methods is investigated. And the authors find that both trained and un-trained methods are vulnerable to adversarial perturbations.
Abstract: Deep neural networks give state-of-the-art accuracy for reconstructing images from few and noisy measurements, a problem arising for example in accelerated magnetic resonance imaging (MRI). However, recent works have raised concerns that deep-learning-based image reconstruction methods are sensitive to perturbations and are less robust than traditional methods: Neural networks (i) may be sensitive to small, yet adversarially-selected perturbations, (ii) may perform poorly under distribution shifts, and (iii) may fail to recover small but important features in an image. In order to understand the sensitivity to such perturbations, in this work, we measure the robustness of different approaches for image reconstruction including trained and un-trained neural networks as well as traditional sparsity-based methods. We find, contrary to prior works, that both trained and un-trained methods are vulnerable to adversarial perturbations. Moreover, both trained and un-trained methods tuned for a particular dataset suffer very similarly from distribution shifts. Finally, we demonstrate that an image reconstruction method that achieves higher reconstruction quality, also performs better in terms of accurately recovering fine details. Our results indicate that the state-of-the-art deep-learning-based image reconstruction methods provide improved performance than traditional methods without compromising robustness.

Journal ArticleDOI
TL;DR: In this article, the authors reviewed how AI/ML can be applied to improve upstream components of the imaging pipeline, including exam modality selection, hardware design, exam protocol selection, data acquisition, image reconstruction, and image processing.

Journal ArticleDOI
TL;DR: In this article, a 2-dimensional U-Net with varying contraction layers and different convolutional filters was designed to estimate the specific absorption rate (SAR) distribution in realistic body models.
Abstract: The purpose of this study is to investigate feasibility of estimating the specific absorption rate (SAR) in MRI in real time. To this goal, SAR maps are predicted from 3T- and 7T-simulated magnetic resonance (MR) images in 10 realistic human body models via a convolutional neural network. Two-dimensional (2-D) U-Net architectures with varying contraction layers and different convolutional filters were designed to estimate the SAR distribution in realistic body models. Sim4Life (ZMT, Switzerland) was used to create simulated anatomical images and SAR maps at 3T and 7T imaging frequencies for Duke, Ella, Charlie, and Pregnant Women (at 3, 7, and 9 month gestational stages) body models. Mean squared error (MSE) was used as the cost function and the structural similarity index (SSIM) was reported. A 2-D U-Net with 4 contracting (and 4 expanding) layers and 64 convolutional filters at the initial stage showed the best compromise to estimate SAR distributions. Adam optimizer outperformed stochastic gradient descent (SGD) for all cases with an average SSIM of $90.5 \mp 3.6$ % and an average MSE of $0.7 \mp 0.6$ % for head images at 7T, and an SSIM of > $85.1 \mp 6.2$ % and an MSE of $0.4 \mp 0.4$ % for 3T body imaging. Algorithms estimated the SAR maps for $224\times 224$ slices under 30 ms. The proposed methodology shows promise to predict real-time SAR in clinical imaging settings without using extra mapping techniques or patient-specific calibrations.

10 Feb 2021
TL;DR: In this paper, a simple autoencoder and gradient update (Latent Shift) is proposed to transform the latent representation of a specific input image to exaggerate or curtail the features used for prediction.
Abstract: Motivation: Traditional image attribution methods struggle to satisfactorily explain predictions of neural networks. Prediction explanation is important, especially in medical imaging, for avoiding the unintended consequences of deploying AI systems when false positive predictions can impact patient care. Thus, there is a pressing need to develop improved models for model explainability and introspection. Specific problem: A new approach is to transform input images to increase or decrease features which cause the prediction. However, current approaches are difficult to implement as they are monolithic or rely on GANs. These hurdles prevent wide adoption. Our approach: Given an arbitrary classifier, we propose a simple autoencoder and gradient update (Latent Shift) that can transform the latent representation of a specific input image to exaggerate or curtail the features used for prediction. We use this method to study chest X-ray classifiers and evaluate their performance. We conduct a reader study with two radiologists assessing 240 chest X-ray predictions to identify which ones are false positives (half are) using traditional attribution maps or our proposed method. Results: We found low overlap with ground truth pathology masks for models with reasonably high accuracy. However, the results from our reader study indicate that these models are generally looking at the correct features. We also found that the Latent Shift explanation allows a user to have more confidence in true positive predictions compared to traditional approaches (0.15$\pm$0.95 in a 5 point scale with p=0.01) with only a small increase in false positive predictions (0.04$\pm$1.06 with p=0.57). Accompanying webpage: this https URL Source code: this https URL

Journal ArticleDOI
TL;DR: In this article, a fully automated femoral cartilage segmentation model for measuring T2 relaxation values and longitudinal changes using multi-echo spin-echo (MESE) magnetic resonance imaging is presented.
Abstract: ObjectiveWe evaluated a fully automated femoral cartilage segmentation model for measuring T2 relaxation values and longitudinal changes using multi-echo spin-echo (MESE) magnetic resonance imaging...

Posted Content
18 Feb 2021
TL;DR: In this paper, a simple autoencoder and gradient update (Latent Shift) is proposed to transform the latent representation of an input image to exaggerate or curtail the features used for prediction.
Abstract: Motivation: Traditional image attribution methods struggle to satisfactorily explain predictions of neural networks. Prediction explanation is important, especially in the medical imaging, for avoiding the unintended consequences of deploying AI systems when false positive predictions can impact patient care. Thus, there is a pressing need to develop improved models for model explainability and introspection. Specific Problem: A new approach is to transform input images to increase or decrease features which cause the prediction. However, current approaches are difficult to implement as they are monolithic or rely on GANs. These hurdles prevent wide adoption. Our approach: Given an arbitrary classifier, we propose a simple autoencoder and gradient update (Latent Shift) that can transform the latent representation of an input image to exaggerate or curtail the features used for prediction. We use this method to study chest X-ray classifiers and evaluate their performance. We conduct a reader study with two radiologists assessing 240 chest X-ray predictions to identify which ones are false positives (half are) using traditional attribution maps or our proposed method. Results: We found low overlap with ground truth pathology masks for models with reasonably high accuracy. However, the results from our reader study indicate that these models are generally looking at the correct features. We also found that the Latent Shift explanation allows a user to have more confidence in true positive predictions compared to traditional approaches (0.15$\pm$0.95 in a 5 point scale with p=0.01) with only a small increase in false positive predictions (0.04$\pm$1.06 with p=0.57). Accompanying webpage: this https URL Source code: this https URL

Posted Content
TL;DR: In this paper, a semi-supervised, consistency-based framework (termed Noise2Recon) is proposed for joint MR reconstruction and denoising, which enables the usage of a limited number of fully-sampled and a large number of undersampled-only scans.
Abstract: Deep learning (DL) has shown promise for faster, high quality accelerated MRI reconstruction. However, standard supervised DL methods depend on extensive amounts of fully-sampled ground-truth data and are sensitive to out-of-distribution (OOD) shifts, in particular for low signal-to-noise ratio (SNR) acquisitions. To alleviate this challenge, we propose a semi-supervised, consistency-based framework (termed Noise2Recon) for joint MR reconstruction and denoising. Our method enables the usage of a limited number of fully-sampled and a large number of undersampled-only scans. We compare our method to augmentation-based supervised techniques and fine-tuned denoisers. Results demonstrate that even with minimal ground-truth data, Noise2Recon (1) achieves high performance on in-distribution (low-noise) scans and (2) improves generalizability to OOD, noisy scans.

DOI
01 Oct 2021
TL;DR: In this paper, a deep learning system integrated within a hardware doppler platform for low-cost, fast, and accessible assessment of disease at the point-of-care is proposed.
Abstract: The most ubiquitous tool utilized at the point-of-care within clinic for the evaluation of Peripheral Arterial Disease (PAD) or limb perfusion is the hand-held doppler. Providers interpret audible circulatory sounds in order to predict waveform phasicity as Monophasic (severe disease), Biphasic (moderate disease), or Triphasic (normal). This prediction is highly subjective. With any doubt in interpretation, patients are referred to an ultrasound laboratory for formal evaluation which can delay definitive diagnosis and treatment. This paper proposes a novel deep learning system integrated within a hardware doppler platform for low-cost, fast, and accessible assessment of disease at the point-of-care. The system converts input sound into a spectrogram and classifies waveform phasicity with a neural network through learned visual temporal amplitude patterns. The hardware system consists of a low-cost single board computer and simple peripherals including a microphone and LCD display. The final system received a validation accuracy of 96.23%, an F1 score of 90.57%, and was additionally tested within a manually simulated clinical environment with interference in which it received an accuracy of 91.66%. Such a tool can be used to more objectively and efficiently diagnose PAD at the point-of-care and prevent delays in intervention.

Posted Content
TL;DR: In this paper, the authors propose applying physics-driven data augmentations for consistency training that leverage our domain knowledge of the forward MRI data acquisition process and MRI physics for improved data efficiency and robustness to clinically-relevant distribution drifts.
Abstract: Deep neural networks have enabled improved image quality and fast inference times for various inverse problems, including accelerated magnetic resonance imaging (MRI) reconstruction. However, such models require large amounts of fully-sampled ground truth data, which are difficult to curate and are sensitive to distribution drifts. In this work, we propose applying physics-driven data augmentations for consistency training that leverage our domain knowledge of the forward MRI data acquisition process and MRI physics for improved data efficiency and robustness to clinically-relevant distribution drifts. Our approach, termed VORTEX (1) demonstrates strong improvements over supervised baselines with and without augmentation in robustness to signal-to-noise ratio change and motion corruption in data-limited regimes; (2) considerably outperforms state-of-the-art data augmentation techniques that are purely image-based on both in-distribution and out-of-distribution data; and (3) enables composing heterogeneous image-based and physics-driven augmentations.

Posted Content
TL;DR: In this article, a simple autoencoder and gradient update (Latent Shift) is proposed to transform the latent representation of a specific input image to exaggerate or curtail the features used for prediction.
Abstract: Motivation: Traditional image attribution methods struggle to satisfactorily explain predictions of neural networks. Prediction explanation is important, especially in medical imaging, for avoiding the unintended consequences of deploying AI systems when false positive predictions can impact patient care. Thus, there is a pressing need to develop improved models for model explainability and introspection. Specific problem: A new approach is to transform input images to increase or decrease features which cause the prediction. However, current approaches are difficult to implement as they are monolithic or rely on GANs. These hurdles prevent wide adoption. Our approach: Given an arbitrary classifier, we propose a simple autoencoder and gradient update (Latent Shift) that can transform the latent representation of a specific input image to exaggerate or curtail the features used for prediction. We use this method to study chest X-ray classifiers and evaluate their performance. We conduct a reader study with two radiologists assessing 240 chest X-ray predictions to identify which ones are false positives (half are) using traditional attribution maps or our proposed method. Results: We found low overlap with ground truth pathology masks for models with reasonably high accuracy. However, the results from our reader study indicate that these models are generally looking at the correct features. We also found that the Latent Shift explanation allows a user to have more confidence in true positive predictions compared to traditional approaches (0.15$\pm$0.95 in a 5 point scale with p=0.01) with only a small increase in false positive predictions (0.04$\pm$1.06 with p=0.57). Accompanying webpage: this https URL Source code: this https URL


Posted Content
TL;DR: In this article, an artificial intelligence-based exam-integrated real-time MRI safety prediction software (MRSaiFE) was proposed to facilitate the safe generation of 3T and 7T images by means of accurate local SAR-monitoring at sub-W/kg levels.
Abstract: A to-date unsolved and highly limiting safety concern for Ultra High-Field (UHF) magnetic resonance imaging (MRI) is the deposition of radiofrequency (RF) power in the body, quantified by the specific absorption rate (SAR), leading to dangerous tissue heating/damage in the form of local SAR hotspots that cannot currently be measured/monitored, thereby severely limiting the applicability of the technology for clinical practice and in regulatory approval. The goal of this study has been to show proof of concept of an artificial intelligence (AI) based exam-integrated real-time MRI safety prediction software (MRSaiFE) facilitating the safe generation of 3T and 7T images by means of accurate local SAR-monitoring at sub-W/kg levels. We trained the software with a small database of image as a feasibility study and achieved successful proof of concept for both field strengths. SAR patterns were predicted with a residual root mean squared error (RSME) of 84% for both field strengths (3T and 7T).

Posted Content
TL;DR: MedPerf as mentioned in this paper is an open framework for benchmarking machine learning in the medical domain, which enables federated evaluation in which models are securely distributed to different facilities for evaluation, thereby empowering healthcare organizations to assess and verify the performance of AI models in an efficient and human-supervised process.
Abstract: Medical AI has tremendous potential to advance healthcare by supporting the evidence-based practice of medicine, personalizing patient treatment, reducing costs, and improving provider and patient experience. We argue that unlocking this potential requires a systematic way to measure the performance of medical AI models on large-scale heterogeneous data. To meet this need, we are building MedPerf, an open framework for benchmarking machine learning in the medical domain. MedPerf will enable federated evaluation in which models are securely distributed to different facilities for evaluation, thereby empowering healthcare organizations to assess and verify the performance of AI models in an efficient and human-supervised process, while prioritizing privacy. We describe the current challenges healthcare and AI communities face, the need for an open platform, the design philosophy of MedPerf, its current implementation status, and our roadmap. We call for researchers and organizations to join us in creating the MedPerf open benchmarking platform.

Posted Content
TL;DR: Deep Inversion for Synthesizing Counterfactuals (DISC) as mentioned in this paper improves upon deep inversion by utilizing stronger image priors, incorporating a novel manifold consistency objective and adopting a progressive optimization strategy.
Abstract: Explanation techniques that synthesize small, interpretable changes to a given image while producing desired changes in the model prediction have become popular for introspecting black-box models. Commonly referred to as counterfactuals, the synthesized explanations are required to contain discernible changes (for easy interpretability) while also being realistic (consistency to the data manifold). In this paper, we focus on the case where we have access only to the trained deep classifier and not the actual training data. While the problem of inverting deep models to synthesize images from the training distribution has been explored, our goal is to develop a deep inversion approach to generate counterfactual explanations for a given query image. Despite their effectiveness in conditional image synthesis, we show that existing deep inversion methods are insufficient for producing meaningful counterfactuals. We propose DISC (Deep Inversion for Synthesizing Counterfactuals) that improves upon deep inversion by utilizing (a) stronger image priors, (b) incorporating a novel manifold consistency objective and (c) adopting a progressive optimization strategy. We find that, in addition to producing visually meaningful explanations, the counterfactuals from DISC are effective at learning classifier decision boundaries and are robust to unknown test-time corruptions.

Posted Content
TL;DR: OncoNet as discussed by the authors is a machine learning algorithm that assesses treatment response from a 1,954 pairs of sequential FDG PET/CT exams through weak supervision using the standard uptake values (SUVmax) in associated radiology reports.
Abstract: FDG PET/CT imaging is a resource intensive examination critical for managing malignant disease and is particularly important for longitudinal assessment during therapy. Approaches to automate longtudinal analysis present many challenges including lack of available longitudinal datasets, managing complex large multimodal imaging examinations, and need for detailed annotations for traditional supervised machine learning. In this work we develop OncoNet, novel machine learning algorithm that assesses treatment response from a 1,954 pairs of sequential FDG PET/CT exams through weak supervision using the standard uptake values (SUVmax) in associated radiology reports. OncoNet demonstrates an AUROC of 0.86 and 0.84 on internal and external institution test sets respectively for determination of change between scans while also showing strong agreement to clinical scoring systems with a kappa score of 0.8. We also curated a dataset of 1,954 paired FDG PET/CT exams designed for response assessment for the broader machine learning in healthcare research community. Automated assessment of radiographic response from FDG PET/CT with OncoNet could provide clinicians with a valuable tool to rapidly and consistently interpret change over time in longitudinal multi-modal imaging exams.