Background/Objectives: Recent state-of-the-art advances in deep learning have significantly improved diagnostic accuracy in medical imaging, particularly in chest radiograph (CXR) analysis. Motivated by these developments, a comprehensive comparison was conducted to investigate how architectural choices affect performance of 14 deep learning models across Convolutional Neural Networks (CNNs), Transformer-based models, and Mamba-based State Space Models. Methods: These models were trained and evaluated under identical conditions on the NIH ChestX-ray14 dataset, a large-scale and widely used benchmark comprising 112,120 labeled CXR images with 14 thoracic disease categories. Results: It was found that recent hybrid architectures—particularly ConvFormer, CaFormer, and EfficientNet—deliver superior performance in both common and rare pathologies. ConvFormer achieved the highest mean AUROC of 0.841 when averaged across all 14 thoracic disease classes, closely followed by EfficientNet and CaFormer. Notably, AUROC scores of 0.94 for hernia, 0.91 for cardiomegaly, and 0.88 for edema and effusion were achieved by the proposed models, surpassing previously reported benchmarks. Conclusions: These results not only highlight the continued strength of CNNs but also demonstrate the growing potential of Transformer-based architectures in medical image analysis. This work contributes to the literature by providing a unified, state-of-the-art benchmarking of diverse deep learning models, offering valuable guidance for researchers and practitioners developing clinically robust AI systems for radiology.

The purpose of this study is to establish local diagnostic reference levels (LDRLs) for computed tomography (CT) procedures using cloud-based automated dose-tracking software. The study includes the dose data obtained from a total of 104,272 examinations performed on adult patients (>18 years) using 8 CT scanners over 12 months. The protocols included in our study were as follows: head CT without contrast, cervical spine CT without contrast, neck CT with contrast, chest CT without contrast, abdomen-pelvis CT without contrast, lumbar spine CT without contrast, high-resolution computed tomography (HRCT) of the chest, and coronary CT angiography (CTA). Dose data were collected using cloud-based automatic dose-tracking software. The 75^th percentiles of the distributions of the median volume CT dose index (CTDIvol) and dose length product (DLP) values were used to determine the LDRLs for each protocol. The LDRLs were compared with national DRLs (NDRLs) and DRLs set in other countries. Inter-CT scanner variability, which is a measure of how well clinical practices are standardized, was determined for each protocol. Median values for each protocol were compared with the LDRLs for dose optimization in each CT scanner. The LDRLs (for DLP and CTDIvol, respectively) were 839 mGy.cm and 41.2 mGy for head CT without contrast, 530.6 mGy.cm and 19.8 mGy for cervical spine CT without contrast, 431.9 mGy.cm and 15.5 mGy for neck CT with contrast, 364.8 mGy.cm and 9.3 mGy for chest CT without contrast, 588.9 mGy. cm and 11.2 mGy for abdomen-pelvis CT without contrast, 713 mGy.cm and 24.3 mGy for lumbar spine CT without contrast, 326 mGy.cm and 9.5 mGy for HRCT, and 642.3 mGy.cm and 33.4 mGy for coronary CTA. The LDRLs were comparable to or lower than NDRLs and DRLs set in other countries for most protocols. The comparisons revealed the need for immediate initiation of an optimization process for CT protocols with higher dose distributions. Furthermore, protocols with high inter-CT scanner variability revealed the need for standardization. There is a need to update the NDRLs for CT protocols in Turkey. Until new NDRLs are established, local institutions in Turkey can initiate the optimization process by comparing their dose distributions to the LDRLs established in our study. Automated dose-tracking software can play an important role in establishing DRLs by facilitating the collection and analysis of large datasets.

Objective: Named entity recognition (NER) offers a powerful method for automatically extracting key clinical information from text, but current models often lack sufficient support for non-English languages. Materials and Methods: This study investigated a prompt-based NER approach using Google’s Gemini 1.5 Pro, a large language model (LLM) with a 1.5-million-token context window. We focused on extracting important clinical entities from Turkish mammography reports, a language with limited available natural language processing (NLP) tools. Our method employed many-shot learning, incorporating 165 examples within a 26,000-token prompt derived from 75 initial reports. We tested the model on a separate set of 85 unannotated reports, concentrating on five key entities: anatomy (ANAT), impression (IMP), observation presence (OBS-P), absence (OBS-A), and uncertainty (OBS-U). Results: Our approach achieved high accuracy, with a macro-averaged F1 score of 0.99 for relaxed match and 0.84 for exact match. In relaxed matching, the model achieved F1 scores of 0.99 for ANAT, 0.99 for IMP, 1.00 for OBS-P, 1.00 for OBS-A, and 0.99 for OBS-U. For exact match, the F1 scores were 0.88 for ANAT, 0.79 for IMP, 0.78 for OBS-P, 0.94 for OBS-A, and 0.82 for OBS-U. Discussion: These results indicate that a many-shot prompt engineering approach with large language models provides an effective way to automate clinical information extraction for languages where NLP resources are less developed, and as reported in the literature, generally outperforms zero-shot, five-shot, and other few-shot methods. Conclusion: This approach has the potential to significantly improve clinical workflows and research efforts in multilingual healthcare environments.