Medical image analysis最新文献

This study introduces HCC-Net, a novel wavelet-based approach for the accurate diagnosis of hepatocellular carcinoma (HCC) from abdominal ultrasound (US) images using artificial neural networks. The HCC-Net integrates the discrete wavelet transform (DWT) to decompose US images into four sub-band images, a lesion detector for hierarchical lesion localization, and a pattern-augmented classifier for generating pattern-enhanced lesion images and subsequent classification. The lesion detection uses a hierarchical coarse-to-fine approach to minimize missed lesions. CoarseNet performs initial lesion localization, while FineNet identifies any lesions that were missed. In the classification phase, the wavelet components of detected lesions are synthesized to create pattern-augmented images that enhance feature distinction, resulting in highly accurate classifications. These augmented images are classified into 'Normal,' 'Benign,' or 'Malignant' categories according to their morphologic features on sonography. The experimental results demonstrate the significant effectiveness of the proposed coarse-to-fine detection framework and pattern-augmented classifier in lesion detection and classification. We achieved an accuracy of 96.2 %, a sensitivity of 97.6 %, and a specificity of 98.1 % on the Samsung Medical Center dataset, indicating HCC-Net's potential as a reliable tool for liver cancer screening.

Background: Advancements in tomographic medical imaging have revolutionized diagnostics and treatment monitoring by offering detailed 3D visualization of internal structures. Despite the significant value of computed tomography (CT), challenges such as high radiation dosage and cost barriers limit its accessibility, especially in low- and middle-income countries. Recognizing the potential of radiographic imaging in reconstructing CT images, this scoping review aims to explore the emerging field of synthesizing 3D CT-like images from 2D radiographs by examining the current methodologies.

Methods: A scoping review was carried out following PRISMA-SR guidelines. Eligibility criteria for the articles included full-text articles published up to September 9, 2024, studying methodologies for the synthesis of 3D CT images from 2D biplanar or four-projection x-ray images. Eligible articles were sourced from PubMed MEDLINE, Embase, and arXiv.

Results: 76 studies were included. The majority (50.8 %, n = 30) were published between 2010 and 2020 (38.2 %, n = 29) and from 2020 onwards (36.8 %, n = 28), with European (40.8 %, n = 31), North American (26.3 %, n = 20), and Asian (32.9 %, n = 25) institutions being primary contributors. Anatomical regions varied, with 17.1 % (n = 13) of studies not using clinical data. Further, studies focused on the chest (25 %, n = 19), spine and vertebrae (17.1 %, n = 13), coronary arteries (10.5 %, n = 8), and cranial structures (10.5 %, n = 8), among other anatomical regions. Convolutional neural networks (CNN) (19.7 %, n = 15), generative adversarial networks (21.1 %, n = 16) and statistical shape models (15.8 %, n = 12) emerged as the most applied methodologies. A limited number of studies included explored the use of conditional diffusion models, iterative reconstruction algorithms, statistical shape models, and digital tomosynthesis.

Conclusion: This scoping review summarizes current strategies and challenges in synthetic imaging generation. The development of 3D CT-like imaging from 2D radiographs could reduce radiation risk while simultaneously addressing financial and logistical obstacles that impede global access to CT imaging. Despite initial promising results, the field encounters challenges with varied methodologies and frequent lack of proper validation, requiring further research to define synthetic imaging's clinical role.

Cardiovascular diseases (CVD) can be diagnosed using various diagnostic modalities. The electrocardiogram (ECG) is a cost-effective and widely available diagnostic aid that provides functional information of the heart. However, its ability to classify and spatially localise CVD is limited. In contrast, cardiac magnetic resonance (CMR) imaging provides detailed structural information of the heart and thus enables evidence-based diagnosis of CVD, but long scan times and high costs limit its use in clinical routine. In this work, we present a deep learning strategy for cost-effective and comprehensive cardiac screening solely from ECG. Our approach combines multimodal contrastive learning with masked data modelling to transfer domain-specific information from CMR imaging to ECG representations. In extensive experiments using data from 40,044 UK Biobank subjects, we demonstrate the utility and generalisability of our method for subject-specific risk prediction of CVD and the prediction of cardiac phenotypes using only ECG data. Specifically, our novel multimodal pre-training paradigm improves performance by up to 12.19% for risk prediction and 27.59% for phenotype prediction. In a qualitative analysis, we demonstrate that our learned ECG representations incorporate information from CMR image regions of interest. Our entire pipeline is publicly available at https://github.com/oetu/MMCL-ECG-CMR.

Predicting disease-related molecular traits from histomorphology brings great opportunities for precision medicine. Despite the rich information present in histopathological images, extracting fine-grained molecular features from standard whole slide images (WSI) is non-trivial. The task is further complicated by the lack of annotations for subtyping and contextual histomorphological features that might span multiple scales. This work proposes a novel multiple-instance learning (MIL) framework capable of WSI-based cancer morpho-molecular subtyping by fusion of different-scale features. Our method, debuting as Inter-MIL, follows a weakly-supervised scheme. It enables the training of the patch-level encoder for WSI in a task-aware optimisation procedure, a step normally not modelled in most existing MIL-based WSI analysis frameworks. We demonstrate that optimising the patch-level encoder is crucial to achieving high-quality fine-grained and tissue-level subtyping results and offers a significant improvement over task-agnostic encoders. Our approach deploys a pseudo-label propagation strategy to update the patch encoder iteratively, allowing discriminative subtype features to be learned. This mechanism also empowers extracting fine-grained attention within image tiles (the small patches), a task largely ignored in most existing weakly supervised-based frameworks. With Inter-MIL, we carried out four challenging cancer molecular subtyping tasks in the context of ovarian, colorectal, lung, and breast cancer. Extensive evaluation results show that Inter-MIL is a robust framework for cancer morpho-molecular subtyping with superior performance compared to several recently proposed methods, in small dataset scenarios where the number of available training slides is less than 100. The iterative optimisation mechanism of Inter-MIL significantly improves the quality of the image features learned by the patch embedded and generally directs the attention map to areas that better align with experts' interpretation, leading to the identification of more reliable histopathology biomarkers. Moreover, an external validation cohort is used to verify the robustness of Inter-MIL on molecular trait prediction.

Deep learning methods have been widely used for various glioma predictions. However, they are usually task-specific, segmentation-dependent and lack of interpretable biomarkers. How to accurately predict the glioma histological grade and molecular subtypes at the same time and provide reliable imaging biomarkers is still challenging. To achieve this, we propose a novel cooperative multi-task learning network (CMTLNet) which consists of a task-common feature extraction (CFE) module, a task-specific unique feature extraction (UFE) module and a unique-common feature collaborative classification (UCFC) module. In CFE, a segmentation-free tumor feature perception (SFTFP) module is first designed to extract the tumor-aware features in a classification manner rather than a segmentation manner. Following that, based on the multi-scale tumor-aware features extracted by SFTFP module, CFE uses convolutional layers to further refine these features, from which the task-common features are learned. In UFE, based on orthogonal projection and conditional classification strategies, the task-specific unique features are extracted. In UCFC, the unique and common features are fused with an attention mechanism to make them adaptive to different glioma prediction tasks. Finally, deep features-guided interpretable radiomic biomarkers for each glioma prediction task are explored by combining SHAP values and correlation analysis. Through the comparisons with recent reported methods on a large multi-center dataset comprising over 1800 cases, we demonstrated the superiority of the proposed CMTLNet, with the mean Matthews correlation coefficient in validation and test sets improved by (4.1%, 10.7%), (3.6%, 23.4%), and (2.7%, 22.7%) respectively for glioma grading, 1p/19q and IDH status prediction tasks. In addition, we found that some radiomic features are highly related to uninterpretable deep features and that their variation trends are consistent in multi-center datasets, which can be taken as reliable imaging biomarkers for glioma diagnosis. The proposed CMTLNet provides an interpretable tool for glioma multi-task prediction, which is beneficial for glioma precise diagnosis and personalized treatment.

Transthoracic Echocardiography (TTE) is a crucial tool for assessing cardiac morphology and function quickly and non-invasively without ionising radiation. However, the examination is subject to intra- and inter-user variability and recordings are often limited to 2D imaging and assessments of end-diastolic and end-systolic volumes. We have developed a novel, fully automated machine learning-based framework to generate a personalised 4D (3D plus time) model of the left ventricular (LV) blood pool with high temporal resolution. A 4D shape is reconstructed from specific 2D echocardiographic views employing deep neural networks, pretrained on a synthetic dataset, and fine-tuned in a self-supervised manner using a novel optimisation method for cross-sectional imaging data. No 3D ground truth is needed for model training. The generated digital twins enhance the interpretation of TTE data by providing a versatile tool for automated analysis of LV volume changes, localisation of infarct areas, and identification of new and clinically relevant biomarkers. Experiments are performed on a multicentre dataset that includes TTE exams of 144 patients with normal TTE and 314 patients with acute myocardial infarction (AMI). The novel biomarkers show a high predictive value for survival (area under the curve (AUC) of 0.82 for 1-year all-cause mortality), demonstrating that personalised 3D shape modelling has the potential to improve diagnostic accuracy and risk assessment.