Automatic radiology report generation can enhance diagnostic efficiency and accuracy. However, clean open-source imaging scan-report pairs are limited in scale and variety. Moreover, the vast amount of radiological texts available online is often too noisy to be directly employed. To address this challenge, we introduce a novel task called Noisy Report Refinement (NRR), which generates radiology reports from noisy free-texts. To achieve this, we propose a report refinement pipeline that leverages large language models (LLMs) enhanced with guided self-critique and report selection strategies. To address the inability of existing radiology report generation metrics in measuring cleanliness, radiological usefulness, and factual correctness across various modalities of reports in NRR task, we introduce a new benchmark, NRRBench, for NRR evaluation. This benchmark includes two online-sourced datasets and four clinically explainable LLM-based metrics: two metrics evaluate the matching rate of radiology entities and modality-specific template attributes respectively, one metric assesses report cleanliness, and a combined metric evaluates overall NRR performance. Experiments demonstrate that guided self-critique and report selection strategies significantly improve the quality of refined reports. Additionally, our proposed metrics show a much higher correlation with noisy rate and error count of reports than radiology report generation metrics in evaluating NRR.

Early allograft dysfunction (EAD) significantly affects liver transplantation prognosis. This study evaluated the effectiveness of artificial intelligence (AI)-assisted methods in accurately diagnosing EAD and identifying its causes. The primary metric for assessing the accuracy was the area under the receiver operating characteristic curve (AUC). Accuracy, sensitivity, and specificity were calculated and analyzed to compare the performance of the AI models with each other and with radiologists. EAD classification followed the criteria established by Olthoff et al. A total of 582 liver transplant patients who underwent transplantation between December 2012 and June 2021 were selected. Among these, 117 patients (mean age 33.5 ± 26.5 years, 80 men) were evaluated. The ultrasound parameters, images, and clinical information of patients were extracted from the database to train the AI model. The AUC for the ultrasound-spectrogram fusion network constructed from four ultrasound images and medical data was 0.968 (95%CI: 0.940, 0.991), outperforming radiologists by 30% for all metrics. AI assistance significantly improved diagnostic accuracy, sensitivity, and specificity (P < 0.050) for both experienced and less-experienced physicians. EAD lacks efficient diagnosis and causation analysis methods. The integration of AI and ultrasound enhances diagnostic accuracy and causation analysis. By modeling only images and data related to blood flow, the AI model effectively analyzed patients with EAD caused by abnormal blood supply. Our model can assist radiologists in reducing judgment discrepancies, potentially benefitting patients with EAD in underdeveloped regions. Furthermore, it enables targeted treatment for those with abnormal blood supply.

Neuroimaging in ALS has contributed considerable academic insights in recent years demonstrating genotype-specific topological changes decades before phenoconversion and characterising longitudinal propagation patterns in specific phenotypes. It has elucidated the radiological underpinnings of specific clinical phenomena such as pseudobulbar affect, apathy, behavioural change, spasticity, and language deficits. Academic concepts such as sexual dimorphism, motor reserve, cognitive reserve, adaptive changes, connectivity-based propagation, pathological stages, and compensatory mechanisms have also been evaluated by imaging. The underpinnings of extra-motor manifestations such as cerebellar, sensory, extrapyramidal and cognitive symptoms have been studied by purpose-designed imaging protocols. Clustering approaches have been implemented to uncover radiologically distinct disease subtypes and machine-learning models have been piloted to accurately classify individual patients into relevant diagnostic, phenotypic, and prognostic categories. Prediction models have been developed for survival in symptomatic patients and phenoconversion in asymptomatic mutation carriers. A range of novel imaging modalities have been implemented and 7 Tesla MRI platforms are increasingly being used in ALS studies. Non-ALS MND conditions, such as PLS, SBMA, and SMA, are now also being increasingly studied by quantitative neuroimaging approaches. A unifying theme of recent imaging papers is the departure from describing focal brain changes to focusing on dynamic structural and functional connectivity alterations. Progressive cortico-cortical, cortico-basal, cortico-cerebellar, cortico-bulbar, and cortico-spinal disconnection has been consistently demonstrated by recent studies and recognised as the primary driver of clinical decline. These studies have led the reconceptualisation of ALS as a "network" or "circuitry disease".

To evaluate deep learning (DL)-based models for detecting periapical bone rarefaction (PBRs) in panoramic radiographs (PRs), analyzing their feasibility and performance in dental practice. A search was conducted across seven databases and partial grey literature up to November 15, 2024, using Medical Subject Headings and entry terms related to DL, PBRs, and PRs. Studies assessing DL-based models for detecting and classifying PBRs in conventional PRs were included, while those using non-PR imaging or focusing solely on non-PBR lesions were excluded. Two independent reviewers performed screening, data extraction, and quality assessment using the Quality Assessment of Diagnostic Accuracy Studies-2 tool, with conflicts resolved by a third reviewer. Twelve studies met the inclusion criteria, mostly from Asia (58.3%). The risk of bias was moderate in 10 studies (83.3%) and high in 2 (16.7%). DL models showed moderate to high performance in PBR detection (sensitivity: 26-100%; specificity: 51-100%), with U-NET and YOLO being the most used algorithms. Only one study (8.3%) distinguished Periapical Granuloma from Periapical Cysts, revealing a classification gap. Key challenges included limited generalization due to small datasets, anatomical superimpositions in PRs, and variability in reported metrics, compromising models comparison. This review underscores that DL-based has the potential to become a valuable tool in dental image diagnostics, but it cannot yet be considered a definitive practice. Multicenter collaboration is needed to diversify data and democratize those tools. Standardized performance reporting is critical for fair comparability between different models.

The purpose of this study is to perform image registration and averaging of multiple free-breathing single-shot cardiac images, where the individual images may have a low signal-to-noise ratio (SNR). To address low SNR encountered in single-shot imaging, especially at low field strengths, we propose a fast deep learning (DL)-based image registration method, called Averaging Morph with Edge Detection (AiM-ED). AiM-ED jointly registers multiple noisy source images to a noisy target image and utilizes a noise-robust pre-trained edge detector to define the training loss. We validate AiM-ED using synthetic late gadolinium enhanced (LGE) images from the MR extended cardiac-torso (MRXCAT) phantom and free-breathing single-shot LGE images from healthy subjects (24 slices) and patients (5 slices) under various levels of added noise. Additionally, we demonstrate the clinical feasibility of AiM-ED by applying it to data from patients (6 slices) scanned on a 0.55T scanner. Compared with a traditional energy-minimization-based image registration method and DL-based VoxelMorph, images registered using AiM-ED exhibit higher values of recovery SNR and three perceptual image quality metrics. An ablation study shows the benefit of both jointly processing multiple source images and using an edge map in AiM-ED. For single-shot LGE imaging, AiM-ED outperforms existing image registration methods in terms of image quality. With fast inference, minimal training data requirements, and robust performance at various noise levels, AiM-ED has the potential to benefit single-shot CMR applications.

Coronary artery disease is the main cause of mortality worldwide mandating early detection, appropriate treatment, and follow-up. Noninvasive cardiac imaging techniques allow detection of obstructive coronary heart disease by direct visualization of the arteries or myocardial blood flow reduction. These techniques have made remarkable progress since their introduction, achieving high diagnostic precision. This review aims at evaluating these noninvasive cardiac imaging techniques, rendering a thorough overview of diagnostic decision-making for detection of ischemia. We discuss the latest advances in the field such as computed tomography angiography, single-photon emission tomography, positron emission tomography, and cardiac magnetic resonance; their main advantages and disadvantages, their most appropriate use and prospects. For the review, we analyzed the literature from 2009 to 2024 on noninvasive cardiac imaging in the diagnosis of coronary artery disease. The review included the 78 publications considered most relevant, including landmark trials, review articles and guidelines. The progress in cardiac imaging is anticipated to overcome various limitations such as high costs, radiation exposure, artifacts, and differences in interpretation among observers. It is expected to lead to more automated scanning processes, and with the assistance of artificial intelligence-driven post-processing software, higher accuracy and reproducibility may be attained.

Deep learning networks map input data to output predictions by fitting network parameters using training data. However, applying a trained network to new, unseen inference data resembles an interpolation process, which may lead to inaccurate predictions if the training and inference data distributions differ significantly. This study aims to generally improve the prediction accuracy of deep learning networks on the inference case by bridging the gap between training and inference data. We propose an inference-specific learning strategy to enhance the network learning process without modifying the network structure. By aligning training data to closely match the specific inference data, we generate an inference-specific training dataset, enhancing the network optimization around the inference data point for more accurate predictions. Taking medical image auto-segmentation as an example, we develop an inference-specific auto-segmentation framework consisting of initial segmentation learning, inference-specific training data deformation, and inference-specific segmentation refinement. The framework is evaluated on public abdominal, head-neck, and pancreas CT datasets comprising 30, 42, and 210 cases, respectively, for medical image segmentation. Experimental results show that our method improves the organ-averaged mean Dice by 6.2% (p-value = 0.001), 1.5% (p-value = 0.003), and 3.7% (p-value < 0.001) on the three datasets, respectively, with a more notable increase for difficult-to-segment organs (such as a 21.7% increase for the gallbladder [p-value = 0.004]). By incorporating organ mask-based weak supervision into the training data alignment learning, the inference-specific auto-segmentation accuracy is generally improved compared with the image intensity-based alignment. Besides, a moving-averaged calculation of the inference organ mask during the learning process strengthens both the robustness and accuracy of the final inference segmentation. By leveraging inference data during training, the proposed inference-specific learning strategy consistently improves auto-segmentation accuracy and holds the potential to be broadly applied for enhanced deep learning decision-making.

Accurate quantitative PET imaging in neurological studies requires proper attenuation correction. MRI-guided attenuation correction in PET/MRI remains challenging owing to the lack of direct relationship between MRI intensities and linear attenuation coefficients. This study aims at generating accurate patient-specific synthetic CT volumes, attenuation maps, and attenuation correction factor (ACF) sinograms with continuous values utilizing a combination of machine learning algorithms, image processing techniques, and voxel-based radiomics feature extraction approaches. Brain MR images of ten healthy volunteers were acquired using IR-pointwise encoding time reduction with radial acquisition (IR-PETRA) and VIBE-Dixon techniques. synthetic CT (SCT) images, attenuation maps, and attenuation correction factors (ACFs) were generated using the LightGBM, a fast and accurate machine learning algorithm, from the radiomics-based and image processing-based feature maps of MR images. Additionally, ultra-low-dose CT images of the same volunteers were acquired and served as the standard of reference for evaluation. The SCT images, attenuation maps, and ACF sinograms were assessed using qualitative and quantitative evaluation metrics and compared against their corresponding reference images, attenuation maps, and ACF sinograms. The voxel-wise and volume-wise comparison between synthetic and reference CT images yielded an average mean absolute error of 60.75 ± 8.8 HUs, an average structural similarity index of 0.88 ± 0.02, and an average peak signal-to-noise ratio of 32.83 ± 2.74 dB. Additionally, we compared MRI-based attenuation maps and ACF sinograms with their CT-based counterparts, revealing average normalized mean absolute errors of 1.48% and 1.33%, respectively. Quantitative assessments indicated higher correlations and similarities between LightGBM-synthesized CT and Reference CT images. Moreover, the cross-validation results showed the possibility of producing accurate SCT images, MRI-based attenuation maps, and ACF sinograms. This might spur the implementation of MRI-based attenuation correction on PET/MRI and dedicated brain PET scanners with lower computational time using CPU-based processors.

Accurate assessment of expected survival in melanoma patients is crucial for treatment decisions. This study explores deep learning-based body composition analysis to predict overall survival (OS) using baseline Computed Tomography (CT) scans and identify fully volumetric, prognostic body composition features. A deep learning network segmented baseline abdomen and thorax CTs from a cohort of 495 patients. The Sarcopenia Index (SI), Myosteatosis Fat Index (MFI), and Visceral Fat Index (VFI) were derived and statistically assessed for prognosticating OS. External validation was performed with 428 patients. SI was significantly associated with OS on both CT regions: abdomen (P ≤ 0.0001, HR: 0.36) and thorax (P ≤ 0.0001, HR: 0.27), with lower SI associated with prolonged survival. MFI was also associated with OS on abdomen (P ≤ 0.0001, HR: 1.16) and thorax CTs (P ≤ 0.0001, HR: 1.08), where higher MFI was linked to worse outcomes. Lastly, VFI was associated with OS on abdomen CTs (P ≤ 0.001, HR: 1.90), with higher VFI linked to poor outcomes. External validation replicated these results. SI, MFI, and VFI showed substantial potential as prognostic factors for OS in malignant melanoma patients. This approach leveraged existing CT scans without additional procedural or financial burdens, highlighting the seamless integration of DL-based body composition analysis into standard oncologic staging routines.

According to the updated classification system, human epidermal growth factor receptor 2 (HER2) expression statuses are divided into the following three groups: HER2-over-expression, HER2-low-expression, and HER2-zero-expression. HER2-negative expression was reclassified into HER2-low-expression and HER2-zero-expression. This study aimed to identify three different HER2 expression statuses for breast cancer (BC) patients using PET/CT radiomics and clinicopathological characteristics. A total of 315 BC patients who met the inclusion and exclusion criteria from two institutions were retrospectively included. The patients in institution 1 were divided into the training set and the independent validation set according to the ratio of 7:3, and institution 2 was used as the external validation set. According to the results of pathological examination, all BC patients were divided into HER2-over-expression, HER2-low-expression, and HER2-zero-expression. First, PET/CT radiomic features and clinicopathological features based on each patient were extracted and collected. Second, multiple methods were used to perform feature screening and feature selection. Then, four machine learning classifiers, including logistic regression (LR), k-nearest neighbor (KNN), support vector machine (SVM), and random forest (RF), were constructed to identify HER2-over-expression vs. others, HER2-low-expression vs. others, and HER2-zero-expression vs. others. The receiver operator characteristic (ROC) curve was plotted to measure the model's predictive power. According to the feature screening process, 8, 10, and 2 radiomics features and 2 clinicopathological features were finally selected to construct three prediction models (HER2-over-expression vs. others, HER2-low-expression vs. others, and HER2-zero-expression vs. others). For HER2-over-expression vs. others, the RF model outperformed other models with an AUC value of 0.843 (95%CI: 0.774-0.897), 0.785 (95%CI: 0.665-0.877), and 0.788 (95%CI: 0.708-0.868) in the training set, independent validation set, and external validation set. Concerning HER2-low-expression vs. others, the outperformance of the LR model over other models was identified with an AUC value of 0.783 (95%CI: 0.708-0.846), 0.756 (95%CI: 0.634-0.854), and 0.779 (95%CI: 0.698-0.860) in the training set, independent validation set, and external validation set. Whereas, the KNN model was confirmed as the optimal model to distinguish HER2-zero-expression from others, with an AUC value of 0.929 (95%CI: 0.890-0.958), 0.847 (95%CI: 0.764-0.910), and 0.835 (95%CI: 0.762-0.908) in the training set, independent validation set, and external validation set. Combined PET/CT radiomic models integrating with clinicopathological characteristics are non-invasively predictive of different HER2 statuses of BC patients.