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.

BACKGROUND: Some researchers have questioned whether artificial intelligence (AI) systems maintain their performance when used for women from populations not considered during the development of the system.
OBJECTIVE: To evaluate the impact of transfer learning as a way of improving the generalization of AI systems in the detection of breast cancer.
METHODS: This retrospective case-control Finnish study involved 191 women diagnosed with breast cancer and 191 matched healthy controls. We selected a state-of-the-art AI system for breast cancer detection trained using a large US dataset. The selected baseline system was evaluated in two experimental settings. First, we examined our private Finnish sample as an independent test set that had not been considered in the development of the system (unseen population). Second, the baseline system was retrained to attempt to improve its performance in the unseen population by means of transfer learning. To analyze performance, we used areas under the receiver operating characteristic curve (AUCs) with DeLong\'s test.
RESULTS: Two versions of the baseline system were considered: ImageOnly and Heatmaps. The ImageOnly and Heatmaps versions yielded mean AUC values of 0.82±0.008 and 0.88±0.003 in the US dataset and 0.56 (95% CI=0.50-0.62) and 0.72 (95% CI=0.67-0.77) when evaluated in the unseen population, respectively. The retrained systems achieved AUC values of 0.61 (95% CI=0.55-0.66) and 0.69 (95% CI=0.64-0.75), respectively. There was no statistical difference between the baseline system and the retrained system.
CONCLUSIONS: Transfer learning with a small study sample did not yield a significant improvement in the generalization of the system.

UNASSIGNED: To evaluate the impact of domain adaptation on the performance of language models in predicting five-point Deauville scores on the basis of clinical fluorine 18 fluorodeoxyglucose PET/CT reports.
UNASSIGNED: The authors retrospectively retrieved 4542 text reports and images for fluorodeoxyglucose PET/CT lymphoma examinations from 2008 to 2018 in the University of Wisconsin-Madison institutional clinical imaging database. Of these total reports, 1664 had Deauville scores that were extracted from the reports and served as training labels. The bidirectional encoder representations from transformers (BERT) model and initialized BERT models BioClinicalBERT, RadBERT, and RoBERTa were adapted to the nuclear medicine domain by pretraining using masked language modeling. These domain-adapted models were then compared with the non-domain-adapted versions on the task of five-point Deauville score prediction. The language models were compared against vision models, multimodal vision-language models, and a nuclear medicine physician, with sevenfold Monte Carlo cross-validation. Means and SDs for accuracy are reported, with P values from paired t testing.
UNASSIGNED: Domain adaptation improved the performance of all language models (P = .01). For example, BERT improved from 61.3% ± 2.9 (SD) five-class accuracy to 65.7% ± 2.2 (P = .01) following domain adaptation. Domain-adapted RoBERTa (named DA RoBERTa) performed best, achieving 77.4% ± 3.4 five-class accuracy; this model performed similarly to its multimodal counterpart (named Multimodal DA RoBERTa) (77.2% ± 3.2) and outperformed the best vision-only model (48.1% ± 3.5, P ≤ .001). A physician given the task on a subset of the data had a five-class accuracy of 66%.
UNASSIGNED: Domain adaptation improved the performance of large language models in predicting Deauville scores in PET/CT reports.Keywords Lymphoma, PET, PET/CT, Transfer Learning, Unsupervised Learning, Convolutional Neural Network (CNN), Nuclear Medicine, Deauville, Natural Language Processing, Multimodal Learning, Artificial Intelligence, Machine Learning, Language Modeling Supplemental material is available for this article. © RSNA, 2023See also the commentary by Abajian in this issue.

Due to the high dimensionality and sparsity of the gene expression matrix in single-cell RNA-sequencing (scRNA-seq) data, coupled with significant noise generated by shallow sequencing, it poses a great challenge for cell clustering methods. While numerous computational methods have been proposed, the majority of existing approaches center on processing the target dataset itself. This approach disregards the wealth of knowledge present within other species and batches of scRNA-seq data. In light of this, our paper proposes a novel method named graph-based deep embedding clustering (GDEC) that leverages transfer learning across species and batches. GDEC integrates graph convolutional networks, effectively overcoming the challenges posed by sparse gene expression matrices. Additionally, the incorporation of DEC in GDEC enables the partitioning of cell clusters within a lower-dimensional space, thereby mitigating the adverse effects of noise on clustering outcomes. GDEC constructs a model based on existing scRNA-seq datasets and then applying transfer learning techniques to fine-tune the model using a limited amount of prior knowledge gleaned from the target dataset. This empowers GDEC to adeptly cluster scRNA-seq data cross different species and batches. Through cross-species and cross-batch clustering experiments, we conducted a comparative analysis between GDEC and conventional packages. Furthermore, we implemented GDEC on the scRNA-seq data of uterine fibroids. Compared results obtained from the Seurat package, GDEC unveiled a novel cell type (epithelial cells) and identified a notable number of new pathways among various cell types, thus underscoring the enhanced analytical capabilities of GDEC. Availability and implementation: https://github.com/YuzhiSun/GDEC/tree/main.

Continuous release of image databases with fully or partially identical inner categories dramatically deteriorates the production of autonomous Computer-Aided Diagnostics (CAD) systems for true comprehensive medical diagnostics. The first challenge is the frequent massive bulk release of medical image databases, which often suffer from two common drawbacks: image duplication and corruption. The many subsequent releases of the same data with the same classes or categories come with no clear evidence of success in the concatenation of those identical classes among image databases. This issue stands as a stumbling block in the path of hypothesis-based experiments for the production of a single learning model that can successfully classify all of them correctly. Removing redundant data, enhancing performance, and optimizing energy resources are among the most challenging aspects. In this article, we propose a global data aggregation scale model that incorporates six image databases selected from specific global resources. The proposed valid learner is based on training all the unique patterns within any given data release, thereby creating a unique dataset hypothetically. The Hash MD5 algorithm (MD5) generates a unique hash value for each image, making it suitable for duplication removal. The T-Distributed Stochastic Neighbor Embedding (t-SNE), with a tunable perplexity parameter, can represent data dimensions. Both the Hash MD5 and t-SNE algorithms are applied recursively, producing a balanced and uniform database containing equal samples per category: normal, pneumonia, and Coronavirus Disease of 2019 (COVID-19). We evaluated the performance of all proposed data and the new automated version using the Inception V3 pre-trained model with various evaluation metrics. The performance outcome of the proposed scale model showed more respectable results than traditional data aggregation, achieving a high accuracy of 98.48%, along with high precision, recall, and F1-score. The results have been proved through a statistical t-test, yielding t-values and p-values. It\'s important to emphasize that all t-values are undeniably significant, and the p-values provide irrefutable evidence against the null hypothesis. Furthermore, it\'s noteworthy that the Final dataset outperformed all other datasets across all metric values when diagnosing various lung infections with the same factors.

Transfer Learning (TL) is an approach which has not yet been widely investigated in healthcare, mostly applied in image data. This study outlines a TL pipeline leveraging Individual Case Safety reports (ICSRs) and Electronic Health Records (EHR), applied for the early detection Adverse Drug Reactions (ADR), evaluated using of alopecia and docetaxel on breast cancer patients as use case.

Industrial data scarcity is one of the largest factors holding back the widespread use of machine learning in manufacturing. To overcome this problem, the concept of transfer learning was developed and has received much attention in recent industrial research. This paper focuses on the problem of time series segmentation and presents the first in-depth research on transfer learning for deep learning-based time series segmentation on the industrial use case of end-of-line pump testing. In particular, we investigate whether the performance of deep learning models can be increased by pretraining the network with data from other domains. Three different scenarios are analyzed: source and target data being closely related, source and target data being distantly related, and source and target data being non-related. The results demonstrate that transfer learning can enhance the performance of time series segmentation models with respect to accuracy and training speed. The benefit can be most clearly seen in scenarios where source and training data are closely related and the number of target training data samples is lowest. However, in the scenario of non-related datasets, cases of negative transfer learning were observed as well. Thus, the research emphasizes the potential, but also the challenges, of industrial transfer learning.

UNASSIGNED: To assess whether transfer learning with a bidirectional encoder representations from transformers (BERT) model, pretrained on a clinical corpus, can perform sentence-level anatomic classification of free-text radiology reports, even for anatomic classes with few positive examples.
UNASSIGNED: This retrospective study included radiology reports of patients who underwent whole-body PET/CT imaging from December 2005 to December 2020. Each sentence in these reports (6272 sentences) was labeled by two annotators according to body part (\"brain,\" \"head & neck,\" \"chest,\" \"abdomen,\" \"limbs,\" \"spine,\" or \"others\"). The BERT-based transfer learning approach was compared with two baseline machine learning approaches: bidirectional long short-term memory (BiLSTM) and the count-based method. Area under the precision-recall curve (AUPRC) and area under the receiver operating characteristic curve (AUC) were computed for each approach, and AUCs were compared using the DeLong test.
UNASSIGNED: The BERT-based approach achieved a macro-averaged AUPRC of 0.88 for classification, outperforming the baselines. AUC results for BERT were significantly higher than those of BiLSTM for all classes and those of the count-based method for the \"brain,\" \"chest,\" \"abdomen,\" and \"others\" classes (P values < .025). AUPRC results for BERT were superior to those of baselines even for classes with few labeled training data (brain: BERT, 0.95, BiLSTM, 0.11, count based, 0.41; limbs: BERT, 0.74, BiLSTM, 0.28, count based, 0.46; spine: BERT, 0.82, BiLSTM, 0.53, count based, 0.69).
UNASSIGNED: The BERT-based transfer learning approach outperformed the BiLSTM and count-based approaches in sentence-level anatomic classification of free-text radiology reports, even for anatomic classes with few labeled training data.Keywords: Anatomy, Comparative Studies, Technology Assessment, Transfer Learning Supplemental material is available for this article. © RSNA, 2023.

Extracting relevant information about infectious diseases is an essential task. However, a significant obstacle in supporting public health research is the lack of methods for effectively mining large amounts of health data.
This study aims to use natural language processing (NLP) to extract the key information (clinical factors, social determinants of health) from published cases in the literature.
The proposed framework integrates a data layer for preparing a data cohort from clinical case reports; an NLP layer to find the clinical and demographic-named entities and relations in the texts; and an evaluation layer for benchmarking performance and analysis. The focus of this study is to extract valuable information from COVID-19 case reports.
The named entity recognition implementation in the NLP layer achieves a performance gain of about 1-3% compared to benchmark methods. Furthermore, even without extensive data labeling, the relation extraction method outperforms benchmark methods in terms of accuracy (by 1-8% better). A thorough examination reveals the disease\'s presence and symptoms prevalence in patients.
A similar approach can be generalized to other infectious diseases. It is worthwhile to use prior knowledge acquired through transfer learning when researching other infectious diseases.