Federated learning-based CT liver tumor detection using a teacher‒student SANet with semisupervised learning

Spatiotemporal knowledge teacher–student reinforcement learning to detect liver tumors without contrast agents

Recognition of liver tumors by predicted hyperspectral features based on patient's computed tomography radiomics features

While recent studies on semi-supervised learning have shown remarkable progress in leveraging both labeled and unlabeled data, most of them presume a basic setting of the model is randomly initialized. In this work, we consider semi-supervised learning and transfer learning jointly, leading to a more practical and competitive paradigm that can utilize both powerful pre-trained models from source domain as well as labeled/unlabeled data in the target domain. To better exploit the value of both pre-trained weights and unlabeled target examples, we introduce adaptive consistency regularization that consists of two complementary components: Adaptive Knowledge Consistency (AKC) on the examples between the source and target model, and Adaptive Representation Consistency (ARC) on the target model between labeled and unlabeled examples. Examples involved in the consistency regularization are adaptively selected according to their potential contributions to the target task. We conduct extensive experiments on popular benchmarks including CIFAR-10, CUB-200, and MURA, by fine-tuning the ImageNet pre-trained ResNet-50 model. Results show that our proposed adaptive consistency regularization outperforms state-of-the-art semi-supervised learning techniques such as Pseudo Label, Mean Teacher, and FixMatch. Moreover, our algorithm is orthogonal to existing methods and thus able to gain additional improvements on top of MixMatch and FixMatch. Our code is available at https://github.com/Walleclipse/Semi-Supervised-Transfer-Learning-Paddle.

Semi-supervised learning is known to achieve better generalisation than a model learned solely from labelled data. Therefore, we propose a new method for estimating a pedestrian pose orientation using a soft-target method, which is a type of semi-supervised learning method. Because a convolutional neural network (CNN) based pose orientation estimation requires large numbers of parameters and operations, we apply the teacher–student algorithm to generate a compressed student model with high accuracy and compactness resembling that of the teacher model by combining a deep network with a random forest. After the teacher model is generated using hard target data, the softened outputs (soft-target data) of the teacher model are used for training the student model. Moreover, the orientation of the pedestrian has specific shape patterns, and a wavelet transform is applied to the input image as a pre-processing step owing to its good spatial frequency localisation property and the ability to preserve both the spatial information and gradient information of an image. For a benchmark dataset considering real driving situations based on a single camera, we used the TUD and KITTI datasets. We applied the proposed algorithm to various driving images in the datasets, and the results indicate that its classification performance with regard to the pose orientation is better than that of other state-of-the-art methods based on a CNN. In addition, the computational speed of the proposed student model is faster than that of other deep CNNs owing to the shorter model structure with a smaller number of parameters.

This article gives an overview of the technical possibilities and clinical applications of modern ultrasound. B-mode and Doppler imaging are already essential in daily routine. Advanced technology, such as harmonic imaging, improves image quality. Tissue harmonic imaging provides better resolution by reducing noise, whereas the introduction of contrast agents increases the perceptibility of focal lesions, e.g., liver tumors. Meanwhile the diagnostic accuracy of contrast-enhanced ultrasound imaging is comparable with CT imaging. Other tools, e.g., elastography, ameliorate the detection of liver fibrosis and tumors. Currently, application of three-dimensional and four-dimensional imaging is of low clinical relevance. Panoramic imaging provides images close to CT or MRI.

Combining multiple expert annotations using semi-supervised learning and graph cuts for medical image segmentation

Primary liver tumors constitute one of the most common tumors. These are aggressive tumors with poor survival. Fluorodeoxyglucose (FDG) positron emission tomography/computed tomography (PET/CT), most commonly used functional imaging, shows limited tracer retention and poor tumor to background ratios (TBR). Novel 68Ga-fibroblast-activation-protein inhibitor (FAPI) PET/CT has shown better tracer uptake and detection efficacy in liver tumors. However, most of the available literature is limited to single center studies with limited number of patients. So, we tried to review and analyze the head-to-head comparison of 18F-FDG PET/CT and 68Ga-FAPI PET/CT in evaluation of liver tumors. Literature available on head to head comparison of diagnostic accuracy of 18F-FDG PET/CT and 68Ga-FAPI PET/CT was searched in databases like PubMed, SCOPUS, EMBASE and Google Scholar for published original studies till April 2023. The relevant studies were selected and assessed using the Revised Tool for the Quality Assessment of Diagnostic Accuracy Studies-2 checklist. A random-effect model was used for calculating pooled sensitivity and specificity. They were represented with 95% confidence intervals (95% CI) and demonstrated in Forest plots. I-square statistic was used to assess heterogeneity in the studies. Pooled sensitivity and specificity of FAPI PET/CT and 18F-FDG PET/CT for detection of primary liver tumors was 94.3% (95% CI: 90.6-96.8%); 89.3% (95% CI: 71.8-97.7%) and 56.1% (95% CI: 49.7-62.5%); 96.4% (95% CI: 81.7-99.9%) respectively. Pooled sensitivity for detection of extrahepatic metastatic disease was 92.2% (range: 88.1-100%; 95% CI: 87.8-95.4%) and 72.4% (range: 69.8-76.5; 95% CI: 65.9-78.2%) respectively. Also, the maximum standardized uptake value (SUVmax) and TBR were higher for FAPI PET/CT than 18F-FDG PET/CT in the included studies. Overall, FAPI PET/CT showed higher sensitivity for detection of liver tumors with better SUVmax and TBR than 18F-FDG PET/CT.

This paper presents an approach to detection and segmentation of liver tumors in 3D computed tomography (CT) images. The automatic detection of tumor can be formulized as novelty detection or two-class classification issue. The method can also be used for tumor segmentation, where each voxel is to be assigned with a correct label, either a tumor class or nontumor class. A voxel is represented with a rich feature vector that distinguishes itself from voxels in different classes. A fast learning algorithm Extreme Learning Machine (ELM) is trained as a voxel classifier. In automatic liver tumor detection, we propose and show that ELM can be trained as a one-class classifier with only healthy liver samples in training. It results in a method of tumor detection based on novelty detection. We compare it with two-class ELM. To extract the boundary of a tumor, we adopt the semi-automatic approach by randomly selecting samples in 3D space within a limited region of interest (ROI) for classifier training. Our approach is validated on a group of patients' CT data and the experiment shows good detection and encouraging segmentation results.

To evaluate the diagnostic performance of semi-supervised learning models for aggressive prostate cancer detection on MRI compared to fully supervised models trained with additional expert annotations. We used 1500 MRI scans from the PI-CAI challenge training subset. Positive scans had 220 human and 205 AI-generated annotations. The mtU-Net (proposed teacher-student semi-supervised approach) was compared to supervised (trained using only 220 human annotations) and semi-supervised (trained on human and AI-generated annotations) nnU-Net. The 205 AI-annotated scans were manually annotated, and a fully supervised model was trained. External validation was performed on a newly annotated dataset from the PROMIS study (n = 574,403 lesions) and the Prostate158 dataset (n = 158,126 lesions). Patient-level performance was evaluated using the area under the curve (AUC) and lesion-level detection (overlap > 0.10) using average precision (AP), along with 95% confidence Intervals (in brackets), and the DeLong test to compare AUCs against the supervised and fully supervised models. The fully supervised nnU-Net showed the highest performance on the internal PI-CAI test set (AUC = 0.89 [0.87-0.91], AP = 0.65 [0.60-0.70]) and external validation datasets PROMIS (AUC = 0.68 [0.64-0.72], AP = 0.24 [0.20-0.29]) and Prostate158 (AUC = 0.87 [0.82-0.92], AP = 0.64 [0.56-0.72]), significantly outperforming the supervised baseline (p < 0.05). The proposed semi-supervised mtU-Net demonstrated close external validation performance on PROMIS (AUC = 0.66 [0.62-0.71], AP = 0.20 [0.16-0.25]) and Prostate158 (AUC = 0.86 [0.81-0.92], AP = 0.58 [0.49-0.67]), significantly outperforming the supervised baseline on both datasets (p = 0.047 and p = 0.014, respectively), and showing no significant difference to the fully supervised model (p = 0.199 and p = 0.702, respectively). In prostate MRI tumor detection, fully supervised learning performed best. However, in external validation, the semi-supervised methods demonstrated performance that approached that of the fully supervised model, proving a valuable approach when expert annotations are limited. Question The need for extensive expert voxel-level annotations delays the development of AI-based prostate cancer diagnostic tools and their implementation in clinical practice. Findings The combination of pseudo-labeling with consistency regularization achieved performance comparable to that of fully supervised methods, demonstrating that data diversity matches the impact of expert annotation volume. Clinical relevance Semi-supervised learning reduces dependence on expert annotations while maintaining detection accuracy, enabling the development of scalable, automated diagnostic tools for prostate cancer amid growing clinical workflow demands.

Clinical decision making and research in hepatocellular carcinoma: Pivotal role of imaging techniques

Automatic and efficient liver tumor detection in multi-phase CT images is essential in computer-aided diagnosis of liver tumors. Nowadays, deep learning has been widely used in medical applications. Normally, deep learning-based AI systems need a large quantity of training data, but in the medical field, acquiring sufficient training data with high-quality annotations is a significant challenge. To solve the lack of training data issue, domain adaptation-based methods have recently been developed as a technique to bridge the domain gap across datasets with different feature characteristics and data distributions. This paper presents a domain adaptation-based method for detecting liver tumors in multi-phase CT images. We adopt knowledge for model learning from PV phase images to ART and NC phase images. Clinical Relevance- To minimize the domain gap we employ an adversarial learning scheme with the maximum square loss for mid-level output feature maps using an anchorless detector. Experiments show that our proposed method performs much better for various CT-phase images than normal training.

Automatic image annotation (AIA) has been proved to be an effective and promising solution to automatically deduce the high-level semantics from low-level visual features. Due to the inherent ambiguity of image-label mapping and the scarcity of training examples, it has become a challenge to systematically develop robust annotation models with better performance. In this paper, we try to attack the problem based on 2D CRFs (Conditional Random Fields) and semi-supervised learning which are seamlessly integrated into a unified framework. 2D CRFs can effectively capture the spatial dependency between the neighboring labels, while the semi-supervised learning techniques can exploit the unlabeled data to improve the joint classification performance. We conducted experiments on a medium-sized image collection including about 500 images from Corel Stock Photo CDs. The experimental results demonstrated that the annotation performance of this method outperforms standard CRFs, showing the effectiveness of the proposed unified framework and the feasibility of unlabeled data to help the classification accuracy.

A semi-supervised deep learning approach for cropped image detection

To retrospectively assess the utility of fusion of T2-weighted images (T2WI) and high b-value diffusion-weighted images (DWI) for prostate cancer detection and localization. In this IRB-approved HIPAA-compliant study, 42 patients with prostate cancer underwent MRI including multiplanar T2WI and axial DWI before prostatectomy. Two independent radiologists first assessed multiplanar T2WI and axial DWI(b-1000) images and recorded whether tumor was present in each sextant. Axial T2WI was then fused with axial DWI(b-1000) images, and the radiologists re-evaluated each sextant for tumor. Accuracy was compared using generalized estimating equations based on a binary logistic regression model. The accuracy, sensitivity, specificity, PPV, and NPV for tumor detection on a sextant-basis using separate and fused image sets was 65.1%, 50.8%, 78.0%, 67.8%, and 63.6% and 71.0%, 60.8%, 80.3%, 73.7%, and 69.3%, respectively, for reader 1, and 54.0%, 42.5%, 64.4%, 52.0%, and 55.2%, and 61.1%, 56.7%, 65.2%, 59.6%, and 62.3%, respectively, for reader 2. The improvements in accuracy, sensitivity, and NPV using fused images were statistically significant for both readers, as was the improvement in PPV for reader 2 (P ranging from <0.0001 to 0.041). With either separate or fused images, there was greater sensitivity for tumors of higher grade or larger size (P ranging from <0.001 to 0.099). Fusion of T2WI and high b-value DWI resulted in significant improvements in sensitivity and accuracy for tumor detection on a sextant-basis, with similar specificity.

The uncontrolled proliferation of natural roads in arid regions has exacerbated regional land degradation and desertification, presenting substantial challenges to their accurate mapping owing to their dynamic and obscure features. Moreover, the high cost of data annotation restricts the availability of comprehensively labelled datasets, which are essential for advanced remote sensing processing and natural road detection. This study dedicated to implement a semi-supervised deep learning method for dirty road extraction in southern Mongolia. A new thematic semantic segmentation dataset of natural roads was established firstly to address scarcity of annotation datasets this region. A semi-supervised UniMatch structure was designed consequently. Operating with high-resolution GaoFen-2 images, this approach minimises the need for extensive manual annotation, achieving an IOU of 73.51% and MIOU of 86.37%. This method significantly reduces labour and time costs associated with manual and fully supervised methods. These observations provide a valuable data source and methodology for addressing natural road expansion in arid regions. They can aid governments in evaluating transportation infrastructure in remote areas, and analysing dirty road traffic impact on environment.