Technology offers a lot of potential that is being used to improve the integrity and efficiency of infrastructures. Crack is one of the major concerns that can affect the integrity or usability of any structure. Oftentimes, the use of manual inspection methods leads to delays which can worsen the situation. Automated crack detection has become very necessary for efficient management and inspection of critical infrastructures. Previous research in crack detection employed classification and localization-based models using Deep Convolutional Neural Networks (DCNNs). This study suggests and compares the effectiveness of transfer learned DCNNs for crack detection as a classification model and as a feature extractor to overcome this restriction. The main objective of this paper is to present various methods of crack detection on surfaces and compare their performance over 3 different datasets. Experiments conducted in this work are threefold: initially, the effectiveness of 12 transfer learned DCNN models for crack detection is analyzed on three publicly available datasets: SDNET, CCIC and BSD. With an accuracy of 53.40%, ResNet101 outperformed other models on the SDNET dataset. EfficientNetB0 was the most accurate (98.8%) model on the BSD dataset, and ResNet50 performed better with an accuracy of 99.8% on the CCIC dataset. Secondly, two image enhancement methods are employed to enhance the images and are transferred learned on the 12 DCNNs in pursuance of improving the performance of the SDNET dataset. The results from the experiments show that the enhanced images improved the accuracy of transfer-learned crack detection models significantly. Furthermore, deep features extracted from the last fully connected layer of the DCNNs are used to train the Support Vector Machine (SVM). The integration of deep features with SVM enhanced the detection accuracy across all the DCNN-dataset combinations, according to analysis in terms of accuracy, precision, recall, and F1-score.

BACKGROUND: Large language models (LLMs) have the potential to support promising new applications in health informatics. However, practical data on sample size considerations for fine-tuning LLMs to perform specific tasks in biomedical and health policy contexts are lacking.
OBJECTIVE: This study aims to evaluate sample size and sample selection techniques for fine-tuning LLMs to support improved named entity recognition (NER) for a custom data set of conflicts of interest disclosure statements.
METHODS: A random sample of 200 disclosure statements was prepared for annotation. All \"PERSON\" and \"ORG\" entities were identified by each of the 2 raters, and once appropriate agreement was established, the annotators independently annotated an additional 290 disclosure statements. From the 490 annotated documents, 2500 stratified random samples in different size ranges were drawn. The 2500 training set subsamples were used to fine-tune a selection of language models across 2 model architectures (Bidirectional Encoder Representations from Transformers [BERT] and Generative Pre-trained Transformer [GPT]) for improved NER, and multiple regression was used to assess the relationship between sample size (sentences), entity density (entities per sentence [EPS]), and trained model performance (F1-score). Additionally, single-predictor threshold regression models were used to evaluate the possibility of diminishing marginal returns from increased sample size or entity density.
RESULTS: Fine-tuned models ranged in topline NER performance from F1-score=0.79 to F1-score=0.96 across architectures. Two-predictor multiple linear regression models were statistically significant with multiple R2 ranging from 0.6057 to 0.7896 (all P<.001). EPS and the number of sentences were significant predictors of F1-scores in all cases ( P<.001), except for the GPT-2_large model, where EPS was not a significant predictor (P=.184). Model thresholds indicate points of diminishing marginal return from increased training data set sample size measured by the number of sentences, with point estimates ranging from 439 sentences for RoBERTa_large to 527 sentences for GPT-2_large. Likewise, the threshold regression models indicate a diminishing marginal return for EPS with point estimates between 1.36 and 1.38.
CONCLUSIONS: Relatively modest sample sizes can be used to fine-tune LLMs for NER tasks applied to biomedical text, and training data entity density should representatively approximate entity density in production data. Training data quality and a model architecture\'s intended use (text generation vs text processing or classification) may be as, or more, important as training data volume and model parameter size.

OBJECTIVE: Deep learning can automate delineation in radiation therapy, reducing time and variability. Yet, its efficacy varies across different institutions, scanners, or settings, emphasizing the need for adaptable and robust models in clinical environments. Our study demonstrates the effectiveness of the transfer learning (TL) approach in enhancing the generalizability of deep learning models for auto-segmentation of organs-at-risk (OARs) in cervical brachytherapy.
METHODS: A pre-trained model was developed using 120 scans with ring and tandem applicator on a 3T magnetic resonance (MR) scanner (RT3). Four OARs were segmented and evaluated. Segmentation performance was evaluated by Volumetric Dice Similarity Coefficient (vDSC), 95 % Hausdorff Distance (HD95), surface DSC, and Added Path Length (APL). The model was fine-tuned on three out-of-distribution target groups. Pre- and post-TL outcomes, and influence of number of fine-tuning scans, were compared. A model trained with one group (Single) and a model trained with all four groups (Mixed) were evaluated on both seen and unseen data distributions.
RESULTS: TL enhanced segmentation accuracy across target groups, matching the pre-trained model\'s performance. The first five fine-tuning scans led to the most noticeable improvements, with performance plateauing with more data. TL outperformed training-from-scratch given the same training data. The Mixed model performed similarly to the Single model on RT3 scans but demonstrated superior performance on unseen data.
CONCLUSIONS: TL can improve a model\'s generalizability for OAR segmentation in MR-guided cervical brachytherapy, requiring less fine-tuning data and reduced training time. These results provide a foundation for developing adaptable models to accommodate clinical settings.

Shadow, a natural phenomenon resulting from the absence of light, plays a pivotal role in agriculture, particularly in processes such as photosynthesis in plants. Despite the availability of generic shadow datasets, many suffer from annotation errors and lack detailed representations of agricultural shadows with possible human activity inside, excluding those derived from satellite or drone views. In this paper, we present an evaluation of a synthetically generated top-down shadow segmentation dataset characterized by photorealistic rendering and accurate shadow masks. We aim to determine its efficacy compared to real-world datasets and assess how factors such as annotation quality and image domain influence neural network model training. To establish a baseline, we trained numerous baseline architectures and subsequently explored transfer learning using various freely available shadow datasets. We further evaluated the out-of-domain performance compared to the training set of other shadow datasets. Our findings suggest that AgroSegNet demonstrates competitive performance and is effective for transfer learning, particularly in domains similar to agriculture.

UNASSIGNED: The progress in Colorectal cancer (CRC) screening and management has resulted in an unprecedented caseload for histopathological diagnosis. While artificial intelligence (AI) presents a potential solution, the predominant emphasis on slide-level aggregation performance without thorough verification of cancer in each location, impedes both explainability and transparency. Effectively addressing these challenges is crucial to ensuring the reliability and efficacy of AI in histology applications.
UNASSIGNED: In this study, we created an innovative AI algorithm using transfer learning from a polyp segmentation model in endoscopy. The algorithm precisely localized CRC targets within 0.25 mm² grids from whole slide imaging (WSI). We assessed the CRC detection capabilities at this fine granularity and examined the influence of AI on the diagnostic behavior of pathologists. The evaluation utilized an extensive dataset comprising 858 consecutive patient cases with 1418 WSIs obtained from an external center.
UNASSIGNED: Our results underscore a notable sensitivity of 90.25% and specificity of 96.60% at the grid level, accompanied by a commendable area under the curve (AUC) of 0.962. This translates to an impressive 99.39% sensitivity at the slide level, coupled with a negative likelihood ratio of <0.01, signifying the dependability of the AI system to preclude diagnostic considerations. The positive likelihood ratio of 26.54, surpassing 10 at the grid level, underscores the imperative for meticulous scrutiny of any AI-generated highlights. Consequently, all four participating pathologists demonstrated statistically significant diagnostic improvements with AI assistance.
UNASSIGNED: Our transfer learning approach has successfully yielded an algorithm that can be validated for CRC histological localizations in whole slide imaging. The outcome advocates for the integration of the AI system into histopathological diagnosis, serving either as a diagnostic exclusion application or a computer-aided detection (CADe) tool. This integration has the potential to alleviate the workload of pathologists and ultimately benefit patients.

Deep learning has profoundly influenced various domains, particularly medical image analysis. Traditional transfer learning approaches in this field rely on models pretrained on domain-specific medical datasets, which limits their generalizability and accessibility. In this study, we propose a novel framework called real-world feature transfer learning, which utilizes backbone models initially trained on large-scale general-purpose datasets such as ImageNet. We evaluate the effectiveness and robustness of this approach compared to models trained from scratch, focusing on the task of classifying pneumonia in X-ray images. Our experiments, which included converting grayscale images to RGB format, demonstrate that real-world-feature transfer learning consistently outperforms conventional training approaches across various performance metrics. This advancement has the potential to accelerate deep learning applications in medical imaging by leveraging the rich feature representations learned from general-purpose pretrained models. The proposed methodology overcomes the limitations of domain-specific pretrained models, thereby enabling accelerated innovation in medical diagnostics and healthcare. From a mathematical perspective, we formalize the concept of real-world feature transfer learning and provide a rigorous mathematical formulation of the problem. Our experimental results provide empirical evidence supporting the effectiveness of this approach, laying the foundation for further theoretical analysis and exploration. This work contributes to the broader understanding of feature transferability across domains and has significant implications for the development of accurate and efficient models for medical image analysis, even in resource-constrained settings.

This research focused on the development of a motor imagery (MI) based brain-machine interface (BMI) using deep learning algorithms to control a lower-limb robotic exoskeleton. The study aimed to overcome the limitations of traditional BMI approaches by leveraging the advantages of deep learning, such as automated feature extraction and transfer learning. The experimental protocol to evaluate the BMI was designed as asynchronous, allowing subjects to perform mental tasks at their own will.
A total of five healthy able-bodied subjects were enrolled in this study to participate in a series of experimental sessions. The brain signals from two of these sessions were used to develop a generic deep learning model through transfer learning. Subsequently, this model was fine-tuned during the remaining sessions and subjected to evaluation. Three distinct deep learning approaches were compared: one that did not undergo fine-tuning, another that fine-tuned all layers of the model, and a third one that fine-tuned only the last three layers. The evaluation phase involved the exclusive closed-loop control of the exoskeleton device by the participants\' neural activity using the second deep learning approach for the decoding.
The three deep learning approaches were assessed in comparison to an approach based on spatial features that was trained for each subject and experimental session, demonstrating their superior performance. Interestingly, the deep learning approach without fine-tuning achieved comparable performance to the features-based approach, indicating that a generic model trained on data from different individuals and previous sessions can yield similar efficacy. Among the three deep learning approaches compared, fine-tuning all layer weights demonstrated the highest performance.
This research represents an initial stride toward future calibration-free methods. Despite the efforts to diminish calibration time by leveraging data from other subjects, complete elimination proved unattainable. The study\'s discoveries hold notable significance for advancing calibration-free approaches, offering the promise of minimizing the need for training trials. Furthermore, the experimental evaluation protocol employed in this study aimed to replicate real-life scenarios, granting participants a higher degree of autonomy in decision-making regarding actions such as walking or stopping gait.

We provide in this paper a comprehensive comparison of various transfer learning strategies and deep learning architectures for computer-aided classification of adult-type diffuse gliomas. We evaluate the generalizability of out-of-domain ImageNet representations for a target domain of histopathological images, and study the impact of in-domain adaptation using self-supervised and multi-task learning approaches for pretraining the models using the medium-to-large scale datasets of histopathological images. A semi-supervised learning approach is furthermore proposed, where the fine-tuned models are utilized to predict the labels of unannotated regions of the whole slide images (WSI). The models are subsequently retrained using the ground-truth labels and weak labels determined in the previous step, providing superior performance in comparison to standard in-domain transfer learning with balanced accuracy of 96.91% and F1-score 97.07%, and minimizing the pathologist\'s efforts for annotation. Finally, we provide a visualization tool working at WSI level which generates heatmaps that highlight tumor areas; thus, providing insights to pathologists concerning the most informative parts of the WSI.

UNASSIGNED: X-ray scatter significantly affects the image quality of cone beam computed tomography (CBCT). Although convolutional neural networks (CNNs) have shown promise in correcting x-ray scatter, their effectiveness is hindered by two main challenges: the necessity for extensive datasets and the uncertainty regarding model generalizability. This study introduces a task-based paradigm to overcome these obstacles, enhancing the application of CNNs in scatter correction.
UNASSIGNED: Using a CNN with U-net architecture, the proposed methodology employs a two-stage training process for scatter correction in CBCT scans. Initially, the CNN is pre-trained on approximately 4000 image pairs from geometric phantom projections, then fine-tuned using transfer learning (TL) on 250 image pairs of anthropomorphic projections, enabling task-specific adaptations with minimal data. 2D scatter ratio (SR) maps from projection data were considered as CNN targets, and such maps were used to perform the scatter prediction. The fine-tuning process for specific imaging tasks, like head and neck imaging, involved simulating scans of an anthropomorphic phantom and pre-processing the data for CNN retraining.
UNASSIGNED: For the pre-training stage, it was observed that SR predictions were quite accurate (SSIM≥0.9). The accuracy of SR predictions was further improved after TL, with a relatively short retraining time (≈70 times faster than pre-training) and using considerably fewer samples compared to the pre-training dataset (≈12 times smaller).
UNASSIGNED: A fast and low-cost methodology to generate task-specific CNN for scatter correction in CBCT was developed. CNN models trained with the proposed methodology were successful to correct x-ray scatter in anthropomorphic structures, unknown to the network, for simulated data.

This study aims to establish an effective benign and malignant classification model for breast tumor ultrasound images by using conventional radiomics and transfer learning features. We collaborated with a local hospital and collected a base dataset (Dataset A) consisting of 1050 cases of single lesion 2D ultrasound images from patients, with a total of 593 benign and 357 malignant tumor cases. The experimental approach comprises three main parts: conventional radiomics, transfer learning, and feature fusion. Furthermore, we assessed the model\'s generalizability by utilizing multicenter data obtained from Datasets B and C. The results from conventional radiomics indicated that the SVM classifier achieved the highest balanced accuracy of 0.791, while XGBoost obtained the highest AUC of 0.854. For transfer learning, we extracted deep features from ResNet50, Inception-v3, DenseNet121, MNASNet, and MobileNet. Among these models, MNASNet, with 640-dimensional deep features, yielded the optimal performance, with a balanced accuracy of 0.866, AUC of 0.937, sensitivity of 0.819, and specificity of 0.913. In the feature fusion phase, we trained SVM, ExtraTrees, XGBoost, and LightGBM with early fusion features and evaluated them with weighted voting. This approach achieved the highest balanced accuracy of 0.964 and AUC of 0.981. Combining conventional radiomics and transfer learning features demonstrated clear advantages over using individual features for breast tumor ultrasound image classification. This automated diagnostic model can ease patient burden and provide additional diagnostic support to radiologists. The performance of this model encourages future prospective research in this domain.