Background and Objectives: Malignant bone tumors represent a major problem due to their aggressiveness and low survival rate. One of the determining factors for improving vital and functional prognosis is the shortening of the time between the onset of symptoms and the moment when treatment starts. The objective of the study is to predict the malignancy of a bone tumor from magnetic resonance imaging (MRI) using deep learning algorithms. Materials and Methods: The cohort contained 23 patients in the study (14 women and 9 men with ages between 15 and 80). Two pretrained ResNet50 image classifiers are used to classify T1 and T2 weighted MRI scans. To predict the malignancy of a tumor, a clinical model is used. The model is a feed forward neural network whose inputs are patient clinical data and the output values of T1 and T2 classifiers. Results: For the training step, the accuracies of 93.67% for the T1 classifier and 86.67% for the T2 classifier were obtained. In validation, both classifiers obtained 95.00% accuracy. The clinical model had an accuracy of 80.84% for training phase and 80.56% for validation. The receiver operating characteristic curve (ROC) of the clinical model shows that the algorithm can perform class separation. Conclusions: The proposed method is based on pretrained deep learning classifiers which do not require a manual segmentation of the MRI images. These algorithms can be used to predict the malignancy of a tumor and on the other hand can shorten the time of their diagnosis and treatment process. While the proposed method requires minimal intervention from an imagist, it needs to be tested on a larger cohort of patients.

(1) Background: Coronavirus disease 2019 (COVID-19) is an infectious disease caused by SARS-CoV-2. Reverse transcription polymerase chain reaction (RT-PCR) remains the current gold standard for detecting SARS-CoV-2 infections in nasopharyngeal swabs. In Romania, the first reported patient to have contracted COVID-19 was officially declared on 26 February 2020. (2) Methods: This study proposes a federated learning approach with pre-trained deep learning models for COVID-19 detection. Three clients were locally deployed with their own dataset. The goal of the clients was to collaborate in order to obtain a global model without sharing samples from the dataset. The algorithm we developed was connected to our internal picture archiving and communication system and, after running backwards, it encountered chest CT changes suggestive for COVID-19 in a patient investigated in our medical imaging department on the 28 January 2020. (4) Conclusions: Based on our results, we recommend using an automated AI-assisted software in order to detect COVID-19 based on the lung imaging changes as an adjuvant diagnostic method to the current gold standard (RT-PCR) in order to greatly enhance the management of these patients and also limit the spread of the disease, not only to the general population but also to healthcare professionals.

The integration of deep learning into radiology has the potential to enhance diagnostic processes, yet its acceptance in clinical practice remains limited due to various challenges. This study aimed to develop and evaluate a fine-tuned large language model (LLM), based on Llama 3-8B, to automate the generation of accurate and concise conclusions in magnetic resonance imaging (MRI) and computed tomography (CT) radiology reports, thereby assisting radiologists and improving reporting efficiency. A dataset comprising 15,000 radiology reports was collected from the University of Medicine and Pharmacy of Craiova’s Imaging Center, covering a diverse range of MRI and CT examinations made by four experienced radiologists. The Llama 3-8B model was fine-tuned using transfer-learning techniques, incorporating parameter quantization to 4-bit precision and low-rank adaptation (LoRA) with a rank of 16 to optimize computational efficiency on consumer-grade GPUs. The model was trained over five epochs using an NVIDIA RTX 3090 GPU, with intermediary checkpoints saved for monitoring. Performance was evaluated quantitatively using Bidirectional Encoder Representations from Transformers Score (BERTScore), Recall-Oriented Understudy for Gisting Evaluation (ROUGE), Bilingual Evaluation Understudy (BLEU), and Metric for Evaluation of Translation with Explicit Ordering (METEOR) metrics on a held-out test set. Additionally, a qualitative assessment was conducted, involving 13 independent radiologists who participated in a Turing-like test and provided ratings for the AI-generated conclusions. The fine-tuned model demonstrated strong quantitative performance, achieving a BERTScore F1 of 0.8054, a ROUGE-1 F1 of 0.4998, a ROUGE-L F1 of 0.4628, and a METEOR score of 0.4282. In the human evaluation, the artificial intelligence (AI)-generated conclusions were preferred over human-written ones in approximately 21.8% of cases, indicating that the model’s outputs were competitive with those of experienced radiologists. The average rating of the AI-generated conclusions was 3.65 out of 5, reflecting a generally favorable assessment. Notably, the model maintained its consistency across various types of reports and demonstrated the ability to generalize to unseen data. The fine-tuned Llama 3-8B model effectively generates accurate and coherent conclusions for MRI and CT radiology reports. By automating the conclusion-writing process, this approach can assist radiologists in reducing their workload and enhancing report consistency, potentially addressing some barriers to the adoption of deep learning in clinical practice. The positive evaluations from independent radiologists underscore the model’s potential utility. While the model demonstrated strong performance, limitations such as dataset bias, limited sample diversity, a lack of clinical judgment, and the need for large computational resources require further refinement and real-world validation. Future work should explore the integration of such models into clinical workflows, address ethical and legal considerations, and extend this approach to generate complete radiology reports.

Background: The elimination of the Hepatitis C virus (HCV) will only be possible if rapid and efficient actions are taken. Artificial neural networks (ANNs) are computing systems based on the topology of the biological brain, containing connected artificial neurons that can be tasked with solving medical problems. Aim: We expanded the previously presented HCV micro-elimination project started in September 2020 that aimed to identify HCV infection through coordinated screening in asymptomatic populations and developed two ANN models able to identify at-risk subjects selected through a targeted questionnaire. Material and method: Our study included 14,042 screened participants from a southwestern region of Oltenia, Romania. Each participant completed a 12-item questionnaire along with anti-HCV antibody rapid testing. Hepatitis-C-positive subjects were linked to care and ultimately could receive antiviral treatment if they had detectable viremia. We built two ANNs, trained and tested on the dataset derived from the questionnaires and then used to identify patients in a similar, already existing dataset. Results: We found 114 HCV-positive patients (81 females), resulting in an overall prevalence of 0.81%. We identified sharing personal hygiene items, receiving blood transfusions, having dental work or surgery and re-using hypodermic needles as significant risk factors. When used on an existing dataset of 15,140 persons (119 HCV cases), the first ANN models correctly identified 97 (81.51%) HCV-positive subjects through 13,401 tests, while the second ANN model identified 81 (68.06%) patients through only 5192 tests. Conclusions: The use of ANNs in selecting screening candidates may improve resource allocation and prioritize cases more prone to severe disease.

Contrast-enhanced ultrasound (CEUS) is widely used in the characterization of liver tumors; however, the evaluation of perfusion patterns using CEUS has a subjective character. This study aims to evaluate the accuracy of an automated method based on CEUS for classifying liver lesions and to compare its performance with that of two experienced clinicians. The system used for automatic classification is based on artificial intelligence (AI) algorithms. For an interpretation close to the clinical setting, both clinicians knew which patients were at high risk for hepatocellular carcinoma (HCC), but only one was aware of all the clinical data. In total, 49 patients with 59 liver tumors were included. For the benign and malignant classification, the AI model outperformed both clinicians in terms of specificity (100% vs. 93.33%); still, the sensitivity was lower (74% vs. 93.18% vs. 90.91%). In the second stage of multiclass diagnosis, the automatic model achieved a diagnostic accuracy of 69.93% for HCC and 89.15% for liver metastases. Readers demonstrated greater diagnostic accuracy for HCC (83.05% and 79.66%) and liver metastases (94.92% and 96.61%) compared to the AI system; however, both were experienced sonographers. The AI model could potentially assist and guide less-experienced clinicians to discriminate malignant from benign liver tumors with high accuracy and specificity.

Background: The ultrasound is one of the most used medical imaging investigations worldwide. It is non-invasive and effective in assessing liver tumors or other types of parenchymal changes. Methods: The aim of the study was to build a deep learning model for image segmentation in ultrasound video investigations. The dataset used in the study was provided by the University of Medicine and Pharmacy Craiova, Romania and contained 50 video examinations from 49 patients. The mean age of the patients in the cohort was 69.57. Regarding presence of a subjacent liver disease, 36.73% had liver cirrhosis and 16.32% had chronic viral hepatitis (5 patients: chronic hepatitis C and 3 patients: chronic hepatitis B). Frames were extracted and cropped from each examination and an expert gastroenterologist labelled the lesions in each frame. After labelling, the labels were exported as binary images. A deep learning segmentation model (U-Net) was trained with focal Tversky loss as a loss function. Two models were obtained with two different sets of parameters for the loss function. The performance metrics observed were intersection over union and recall and precision. Results: Analyzing the intersection over union metric, the first segmentation model obtained performed better compared to the second model: 0.8392 (model 1) vs. 0.7990 (model 2). The inference time for both models was between 32.15 milliseconds and 77.59 milliseconds. Conclusions: Two segmentation models were obtained in the study. The models performed similarly during training and validation. However, one model was trained to focus on hard-to-predict labels. The proposed segmentation models can represent a first step in automatically extracting time-intensity curves from CEUS examinations.