Search

Radiology reporting is a complex task that requires detailed image understanding, integration of multiple inputs, including comparison with prior imaging, and precise language generation. This makes it ideal for the development and use of generative multimodal models. Here, we extend report generation to include the localisation of individual findings on the image - a task we call grounded report generation. Prior work indicates that grounding is important for clarifying image understanding and interpreting AI-generated text. Therefore, grounded reporting stands to improve the utility and transparency of automated report drafting. To enable evaluation of grounded reporting, we propose a novel evaluation framework - RadFact - leveraging the reasoning capabilities of large language models (LLMs). RadFact assesses the factuality of individual generated sentences, as well as correctness of generated spatial localisations when present. We introduce MAIRA-2, a large multimodal model combining a radiology-specific image encoder with a LLM, and trained for the new task of grounded report generation on chest X-rays. MAIRA-2 uses more comprehensive inputs than explored previously: the current frontal image, the current lateral image, the prior frontal image and prior report, as well as the Indication, Technique and Comparison sections of the current report. We demonstrate that these additions significantly improve report quality and reduce hallucinations, establishing a new state of the art on findings generation (without grounding) on MIMIC-CXR while demonstrating the feasibility of grounded reporting as a novel and richer task., Comment: 44 pages, 20 figures

Language-supervised pre-training has proven to be a valuable method for extracting semantically meaningful features from images, serving as a foundational element in multimodal systems within the computer vision and medical imaging domains. However, resulting features are limited by the information contained within the text. This is particularly problematic in medical imaging, where radiologists' written findings focus on specific observations; a challenge compounded by the scarcity of paired imaging-text data due to concerns over leakage of personal health information. In this work, we fundamentally challenge the prevailing reliance on language supervision for learning general purpose biomedical imaging encoders. We introduce RAD-DINO, a biomedical image encoder pre-trained solely on unimodal biomedical imaging data that obtains similar or greater performance than state-of-the-art biomedical language supervised models on a diverse range of benchmarks. Specifically, the quality of learned representations is evaluated on standard imaging tasks (classification and semantic segmentation), and a vision-language alignment task (text report generation from images). To further demonstrate the drawback of language supervision, we show that features from RAD-DINO correlate with other medical records (e.g., sex or age) better than language-supervised models, which are generally not mentioned in radiology reports. Finally, we conduct a series of ablations determining the factors in RAD-DINO's performance; notably, we observe that RAD-DINO's downstream performance scales well with the quantity and diversity of training data, demonstrating that image-only supervision is a scalable approach for training a foundational biomedical image encoder.

Biomedical imaging datasets are often small and biased, meaning that real-world performance of predictive models can be substantially lower than expected from internal testing. This work proposes using generative image editing to simulate dataset shifts and diagnose failure modes of biomedical vision models; this can be used in advance of deployment to assess readiness, potentially reducing cost and patient harm. Existing editing methods can produce undesirable changes, with spurious correlations learned due to the co-occurrence of disease and treatment interventions, limiting practical applicability. To address this, we train a text-to-image diffusion model on multiple chest X-ray datasets and introduce a new editing method RadEdit that uses multiple masks, if present, to constrain changes and ensure consistency in the edited images. We consider three types of dataset shifts: acquisition shift, manifestation shift, and population shift, and demonstrate that our approach can diagnose failures and quantify model robustness without additional data collection, complementing more qualitative tools for explainable AI.

We present a radiology-specific multimodal model for the task for generating radiological reports from chest X-rays (CXRs). Our work builds on the idea that large language model(s) can be equipped with multimodal capabilities through alignment with pre-trained vision encoders. On natural images, this has been shown to allow multimodal models to gain image understanding and description capabilities. Our proposed model (MAIRA-1) leverages a CXR-specific image encoder in conjunction with a fine-tuned large language model based on Vicuna-7B, and text-based data augmentation, to produce reports with state-of-the-art quality. In particular, MAIRA-1 significantly improves on the radiologist-aligned RadCliQ metric and across all lexical metrics considered. Manual review of model outputs demonstrates promising fluency and accuracy of generated reports while uncovering failure modes not captured by existing evaluation practices. More information and resources can be found on the project website: https://aka.ms/maira., Comment: 18 pages, 9 tables, 5 figures. v2 adds test IDs and image encoder citation. v3 fixes error in NPV/specificity

In recent years, the field of autonomous driving has witnessed remarkable advancements, driven by the integration of a multitude of sensors, including cameras and LiDAR systems, in different prototypes. However, with the proliferation of sensor data comes the pressing need for more sophisticated information processing techniques. This research paper introduces a novel modification to an object detection network that uses camera and lidar information, incorporating an additional branch designed for the task of re-identifying objects across adjacent cameras within the same vehicle while elevating the quality of the baseline 3D object detection outcomes. The proposed methodology employs a two-step detection pipeline: initially, an object detection network is employed, followed by a 3D box estimator that operates on the filtered point cloud generated from the network's detections. Extensive experimental evaluations encompassing both 2D and 3D domains validate the effectiveness of the proposed approach and the results underscore the superiority of this method over traditional Non-Maximum Suppression (NMS) techniques, with an improvement of more than 5\% in the car category in the overlapping areas.

We introduce a novel Region-based contrastive pretraining for Medical Image Retrieval (RegionMIR) that demonstrates the feasibility of medical image retrieval with similar anatomical regions. RegionMIR addresses two major challenges for medical image retrieval i) standardization of clinically relevant searching criteria (e.g., anatomical, pathology-based), and ii) localization of anatomical area of interests that are semantically meaningful. In this work, we propose an ROI image retrieval image network that retrieves images with similar anatomy by extracting anatomical features (via bounding boxes) and evaluate similarity between pairwise anatomy-categorized features between the query and the database of images using contrastive learning. ROI queries are encoded using a contrastive-pretrained encoder that was fine-tuned for anatomy classification, which generates an anatomical-specific latent space for region-correlated image retrieval. During retrieval, we compare the anatomically encoded query to find similar features within a feature database generated from training samples, and retrieve images with similar regions from training samples. We evaluate our approach on both anatomy classification and image retrieval tasks using the Chest ImaGenome Dataset. Our proposed strategy yields an improvement over state-of-the-art pretraining and co-training strategies, from 92.24 to 94.12 (2.03%) classification accuracy in anatomies. We qualitatively evaluate the image retrieval performance demonstrating generalizability across multiple anatomies with different morphology.

Self-supervised learning in vision-language processing exploits semantic alignment between imaging and text modalities. Prior work in biomedical VLP has mostly relied on the alignment of single image and report pairs even though clinical notes commonly refer to prior images. This does not only introduce poor alignment between the modalities but also a missed opportunity to exploit rich self-supervision through existing temporal content in the data. In this work, we explicitly account for prior images and reports when available during both training and fine-tuning. Our approach, named BioViL-T, uses a CNN-Transformer hybrid multi-image encoder trained jointly with a text model. It is designed to be versatile to arising challenges such as pose variations and missing input images across time. The resulting model excels on downstream tasks both in single- and multi-image setups, achieving state-of-the-art performance on (I) progression classification, (II) phrase grounding, and (III) report generation, whilst offering consistent improvements on disease classification and sentence-similarity tasks. We release a novel multi-modal temporal benchmark dataset, MS-CXR-T, to quantify the quality of vision-language representations in terms of temporal semantics. Our experimental results show the advantages of incorporating prior images and reports to make most use of the data., Comment: To appear in CVPR 2023

The lack of annotated datasets is a major bottleneck for training new task-specific supervised machine learning models, considering that manual annotation is extremely expensive and time-consuming. To address this problem, we present MONAI Label, a free and open-source framework that facilitates the development of applications based on artificial intelligence (AI) models that aim at reducing the time required to annotate radiology datasets. Through MONAI Label, researchers can develop AI annotation applications focusing on their domain of expertise. It allows researchers to readily deploy their apps as services, which can be made available to clinicians via their preferred user interface. Currently, MONAI Label readily supports locally installed (3D Slicer) and web-based (OHIF) frontends and offers two active learning strategies to facilitate and speed up the training of segmentation algorithms. MONAI Label allows researchers to make incremental improvements to their AI-based annotation application by making them available to other researchers and clinicians alike. Additionally, MONAI Label provides sample AI-based interactive and non-interactive labeling applications, that can be used directly off the shelf, as plug-and-play to any given dataset. Significant reduced annotation times using the interactive model can be observed on two public datasets.

Detailed analysis of seizure semiology, the symptoms and signs which occur during a seizure, is critical for management of epilepsy patients. Inter-rater reliability using qualitative visual analysis is often poor for semiological features. Therefore, automatic and quantitative analysis of video-recorded seizures is needed for objective assessment. We present GESTURES, a novel architecture combining convolutional neural networks (CNNs) and recurrent neural networks (RNNs) to learn deep representations of arbitrarily long videos of epileptic seizures. We use a spatiotemporal CNN (STCNN) pre-trained on large human action recognition (HAR) datasets to extract features from short snippets (approx. 0.5 s) sampled from seizure videos. We then train an RNN to learn seizure-level representations from the sequence of features. We curated a dataset of seizure videos from 68 patients and evaluated GESTURES on its ability to classify seizures into focal onset seizures (FOSs) (N = 106) vs. focal to bilateral tonic-clonic seizures (TCSs) (N = 77), obtaining an accuracy of 98.9% using bidirectional long short-term memory (BLSTM) units. We demonstrate that an STCNN trained on a HAR dataset can be used in combination with an RNN to accurately represent arbitrarily long videos of seizures. GESTURES can provide accurate seizure classification by modeling sequences of semiologies., Comment: Accepted at the 24th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI 2021)

Accurate segmentation of brain resection cavities (RCs) aids in postoperative analysis and determining follow-up treatment. Convolutional neural networks (CNNs) are the state-of-the-art image segmentation technique, but require large annotated datasets for training. Annotation of 3D medical images is time-consuming, requires highly-trained raters, and may suffer from high inter-rater variability. Self-supervised learning strategies can leverage unlabeled data for training. We developed an algorithm to simulate resections from preoperative magnetic resonance images (MRIs). We performed self-supervised training of a 3D CNN for RC segmentation using our simulation method. We curated EPISURG, a dataset comprising 430 postoperative and 268 preoperative MRIs from 430 refractory epilepsy patients who underwent resective neurosurgery. We fine-tuned our model on three small annotated datasets from different institutions and on the annotated images in EPISURG, comprising 20, 33, 19 and 133 subjects. The model trained on data with simulated resections obtained median (interquartile range) Dice score coefficients (DSCs) of 81.7 (16.4), 82.4 (36.4), 74.9 (24.2) and 80.5 (18.7) for each of the four datasets. After fine-tuning, DSCs were 89.2 (13.3), 84.1 (19.8), 80.2 (20.1) and 85.2 (10.8). For comparison, inter-rater agreement between human annotators from our previous study was 84.0 (9.9). We present a self-supervised learning strategy for 3D CNNs using simulated RCs to accurately segment real RCs on postoperative MRI. Our method generalizes well to data from different institutions, pathologies and modalities. Source code, segmentation models and the EPISURG dataset are available at https://github.com/fepegar/ressegijcars ., Comment: To be published in the International Journal of Computer Assisted Radiology and Surgery (IJCARS) - Special issue MICCAI 2020

The performance of object detection methods based on LiDAR information is heavily impacted by the availability of training data, usually limited to certain laser devices. As a result, the use of synthetic data is becoming popular when training neural network models, as both sensor specifications and driving scenarios can be generated ad-hoc. However, bridging the gap between virtual and real environments is still an open challenge, as current simulators cannot completely mimic real LiDAR operation. To tackle this issue, domain adaptation strategies are usually applied, obtaining remarkable results on vehicle detection when applied to range view (RV) and bird's eye view (BEV) projections while failing for smaller road agents. In this paper, we present a BEV domain adaptation method based on CycleGAN that uses prior semantic classification in order to preserve the information of small objects of interest during the domain adaptation process. The quality of the generated BEVs has been evaluated using a state-of-the-art 3D object detection framework at KITTI 3D Object Detection Benchmark. The obtained results show the advantages of the proposed method over the existing alternatives., Comment: Submitted to IEEE International Conference on Intelligent Transportation Systems (ITSC2021)

Most sensor setups for onboard autonomous perception are composed of LiDARs and vision systems, as they provide complementary information that improves the reliability of the different algorithms necessary to obtain a robust scene understanding. However, the effective use of information from different sources requires an accurate calibration between the sensors involved, which usually implies a tedious and burdensome process. We present a method to calibrate the extrinsic parameters of any pair of sensors involving LiDARs, monocular or stereo cameras, of the same or different modalities. The procedure is composed of two stages: first, reference points belonging to a custom calibration target are extracted from the data provided by the sensors to be calibrated, and second, the optimal rigid transformation is found through the registration of both point sets. The proposed approach can handle devices with very different resolutions and poses, as usually found in vehicle setups. In order to assess the performance of the proposed method, a novel evaluation suite built on top of a popular simulation framework is introduced. Experiments on the synthetic environment show that our calibration algorithm significantly outperforms existing methods, whereas real data tests corroborate the results obtained in the evaluation suite. Open-source code is available at https://github.com/beltransen/velo2cam_calibration, Comment: Published on IEEE Transactions on Intelligent Transportation Systems

In this paper, a multi-modal 360$^{\circ}$ framework for 3D object detection and tracking for autonomous vehicles is presented. The process is divided into four main stages. First, images are fed into a CNN network to obtain instance segmentation of the surrounding road participants. Second, LiDAR-to-image association is performed for the estimated mask proposals. Then, the isolated points of every object are processed by a PointNet ensemble to compute their corresponding 3D bounding boxes and poses. Lastly, a tracking stage based on Unscented Kalman Filter is used to track the agents along time. The solution, based on a novel sensor fusion configuration, provides accurate and reliable road environment detection. A wide variety of tests of the system, deployed in an autonomous vehicle, have successfully assessed the suitability of the proposed perception stack in a real autonomous driving application., Comment: Accepted for publication in IEEE ITSC 2020

Resective surgery may be curative for drug-resistant focal epilepsy, but only 40% to 70% of patients achieve seizure freedom after surgery. Retrospective quantitative analysis could elucidate patterns in resected structures and patient outcomes to improve resective surgery. However, the resection cavity must first be segmented on the postoperative MR image. Convolutional neural networks (CNNs) are the state-of-the-art image segmentation technique, but require large amounts of annotated data for training. Annotation of medical images is a time-consuming process requiring highly-trained raters, and often suffering from high inter-rater variability. Self-supervised learning can be used to generate training instances from unlabeled data. We developed an algorithm to simulate resections on preoperative MR images. We curated a new dataset, EPISURG, comprising 431 postoperative and 269 preoperative MR images from 431 patients who underwent resective surgery. In addition to EPISURG, we used three public datasets comprising 1813 preoperative MR images for training. We trained a 3D CNN on artificially resected images created on the fly during training, using images from 1) EPISURG, 2) public datasets and 3) both. To evaluate trained models, we calculate Dice score (DSC) between model segmentations and 200 manual annotations performed by three human raters. The model trained on data with manual annotations obtained a median (interquartile range) DSC of 65.3 (30.6). The DSC of our best-performing model, trained with no manual annotations, is 81.7 (14.2). For comparison, inter-rater agreement between human annotators was 84.0 (9.9). We demonstrate a training method for CNNs using simulated resection cavities that can accurately segment real resection cavities, without manual annotations., Comment: 13 pages, 6 figures, accepted at the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI) 2020

Processing of medical images such as MRI or CT presents unique challenges compared to RGB images typically used in computer vision. These include a lack of labels for large datasets, high computational costs, and metadata to describe the physical properties of voxels. Data augmentation is used to artificially increase the size of the training datasets. Training with image patches decreases the need for computational power. Spatial metadata needs to be carefully taken into account in order to ensure a correct alignment of volumes. We present TorchIO, an open-source Python library to enable efficient loading, preprocessing, augmentation and patch-based sampling of medical images for deep learning. TorchIO follows the style of PyTorch and integrates standard medical image processing libraries to efficiently process images during training of neural networks. TorchIO transforms can be composed, reproduced, traced and extended. We provide multiple generic preprocessing and augmentation operations as well as simulation of MRI-specific artifacts. Source code, comprehensive tutorials and extensive documentation for TorchIO can be found at https://torchio.rtfd.io/. The package can be installed from the Python Package Index running 'pip install torchio'. It includes a command-line interface which allows users to apply transforms to image files without using Python. Additionally, we provide a graphical interface within a TorchIO extension in 3D Slicer to visualize the effects of transforms. TorchIO was developed to help researchers standardize medical image processing pipelines and allow them to focus on the deep learning experiments. It encourages open science, as it supports reproducibility and is version controlled so that the software can be cited precisely. Due to its modularity, the library is compatible with other frameworks for deep learning with medical images., Comment: Published in Computer Methods and Programs in Biomedicine

On-board 3D object detection in autonomous vehicles often relies on geometry information captured by LiDAR devices. Albeit image features are typically preferred for detection, numerous approaches take only spatial data as input. Exploiting this information in inference usually involves the use of compact representations such as the Bird's Eye View (BEV) projection, which entails a loss of information and thus hinders the joint inference of all the parameters of the objects' 3D boxes. In this paper, we present a fully end-to-end 3D object detection framework that can infer oriented 3D boxes solely from BEV images by using a two-stage object detector and ad-hoc regression branches, eliminating the need for a post-processing stage. The method outperforms its predecessor (BirdNet) by a large margin and obtains state-of-the-art results on the KITTI 3D Object Detection Benchmark for all the categories in evaluation., Comment: Submitted to IEEE International Conference on Intelligent Transportation Systems (ITSC2020)

The rapid development of embedded hardware in autonomous vehicles broadens their computational capabilities, thus bringing the possibility to mount more complete sensor setups able to handle driving scenarios of higher complexity. As a result, new challenges such as multiple detections of the same object have to be addressed. In this work, a siamese network is integrated into the pipeline of a well-known 3D object detector approach to suppress duplicate proposals coming from different cameras via re-identification. Additionally, associations are exploited to enhance the 3D box regression of the object by aggregating their corresponding LiDAR frustums. The experimental evaluation on the nuScenes dataset shows that the proposed method outperforms traditional NMS approaches., Comment: Submitted to IEEE Intelligent Vehicles Symposium 2020 (IV2020)

Lane detection is extremely important for autonomous vehicles. For this reason, many approaches use lane boundary information to locate the vehicle inside the street, or to integrate GPS-based localization. As many other computer vision based tasks, convolutional neural networks (CNNs) represent the state-of-the-art technology to indentify lane boundaries. However, the position of the lane boundaries w.r.t. the vehicle may not suffice for a reliable positioning, as for path planning or localization information regarding lane types may also be needed. In this work, we present an end-to-end system for lane boundary identification, clustering and classification, based on two cascaded neural networks, that runs in real-time. To build the system, 14336 lane boundaries instances of the TuSimple dataset for lane detection have been labelled using 8 different classes. Our dataset and the code for inference are available online., Comment: Presented at Eurocast 2019

Many systems for autonomous vehicles' navigation rely on lane detection. Traditional algorithms usually estimate only the position of the lanes on the road, but an autonomous control system may also need to know if a lane marking can be crossed or not, and what portion of space inside the lane is free from obstacles, to make safer control decisions. On the other hand, free space detection algorithms only detect navigable areas, without information about lanes. State-of-the-art algorithms use CNNs for both tasks, with significant consumption of computing resources. We propose a novel approach that estimates the free space inside each lane, with a single CNN. Additionally, adding only a small requirement concerning GPU RAM, we infer the road type, that will be useful for path planning. To achieve this result, we train a multi-task CNN. Then, we further elaborate the output of the network, to extract polygons that can be effectively used in navigation control. Finally, we provide a computationally efficient implementation, based on ROS, that can be executed in real time. Our code and trained models are available online., Comment: Will appear in the 2019 IEEE Intelligent Vehicles Symposium (IV 2019)

Understanding driving situations regardless the conditions of the traffic scene is a cornerstone on the path towards autonomous vehicles; however, despite common sensor setups already include complementary devices such as LiDAR or radar, most of the research on perception systems has traditionally focused on computer vision. We present a LiDAR-based 3D object detection pipeline entailing three stages. First, laser information is projected into a novel cell encoding for bird's eye view projection. Later, both object location on the plane and its heading are estimated through a convolutional neural network originally designed for image processing. Finally, 3D oriented detections are computed in a post-processing phase. Experiments on KITTI dataset show that the proposed framework achieves state-of-the-art results among comparable methods. Further tests with different LiDAR sensors in real scenarios assess the multi-device capabilities of the approach., Comment: Submittied to IEEE International Conference on Intelligent Transportation Systems 2018 (ITSC)

Sensor setups consisting of a combination of 3D range scanner lasers and stereo vision systems are becoming a popular choice for on-board perception systems in vehicles; however, the combined use of both sources of information implies a tedious calibration process. We present a method for extrinsic calibration of lidar-stereo camera pairs without user intervention. Our calibration approach is aimed to cope with the constraints commonly found in automotive setups, such as low-resolution and specific sensor poses. To demonstrate the performance of our method, we also introduce a novel approach for the quantitative assessment of the calibration results, based on a simulation environment. Tests using real devices have been conducted as well, proving the usability of the system and the improvement over the existing approaches. Code is available at http://wiki.ros.org/velo2cam_calibration, Comment: Accepted to IEEE International Conference on Intelligent Transportation Systems 2017 (ITSC)