XRaySwinGen: Automatic medical reporting for X-ray exams with multimodal model.

The importance of radiology in modern medicine is acknowledged for its non-invasive diagnostic capabilities, yet the manual formulation of unstructured medical reports poses time constraints and error risks. This study addresses the common limitation of Artificial Intelligence applications in medical image captioning, which typically focus on classification problems, lacking detailed information about the patient's condition. Despite advancements in AI-generated medical reports that incorporate descriptive details from X-ray images, which are essential for comprehensive reports, the challenge persists. The proposed solution involves a multimodal model utilizing Computer Vision for image representation and Natural Language Processing for textual report generation. A notable contribution is the innovative use of the Swin Transformer as the image encoder, enabling hierarchical mapping and enhanced model perception without a surge in parameters or computational costs. The model incorporates GPT-2 as the textual decoder, integrating cross-attention layers and bilingual training with datasets in Portuguese PT-BR and English. Promising results are noted in the proposed database with ROUGE-L 0.748, METEOR 0.741, and NIH CHEST X-ray with ROUGE-L 0.404 and METEOR 0.393.

A ensemble methodology for automatic classification of chest X-rays using deep learning.

Chest radiographies, or chest X-rays, are the most standard imaging exams used in daily hospitals. Responsible for assisting in detecting numerous pathologies and findings that directly interfere in the patient's life, this exam is therefore crucial in screening patients. This work proposes a methodology based on a Convolutional Neural Networks (CNNs) ensemble to aid the diagnosis of chest X-ray exams by screening them with a high probability of being normal or abnormal. In the development of this study, a private dataset with frontal and lateral projections X-ray images was used. To build the ensemble model, VGG-16, ResNet50 and DenseNet121 architectures, which are commonly used in the classification of Chest X-rays, were evaluated. A Confidence Threshold (CTR) was used to define the predictions into High Confidence Normal (HCn), Borderline classification (BC), or High Confidence Abnormal (HCa). In the tests performed, very promising results were achieved: 54.63% of the exams were classified with high confidence; of the normal exams, 32% were classified as HCn with an false discovery rate (FDR) of 1.68%; and as to the abnormal exams, 23% were classified as HCa with 4.91% false omission rate (FOR).

Diagnosis of Leukaemia in Blood Slides Based on a Fine-Tuned and Highly Generalisable Deep Learning Model.

Leukaemia is a dysfunction that affects the production of white blood cells in the bone marrow. Young cells are abnormally produced, replacing normal blood cells. Consequently, the person suffers problems in transporting oxygen and in fighting infections. This article proposes a convolutional neural network (CNN) named LeukNet that was inspired on convolutional blocks of VGG-16, but with smaller dense layers. To define the LeukNet parameters, we evaluated different CNNs models and fine-tuning methods using 18 image datasets, with different resolution, contrast, colour and texture characteristics. We applied data augmentation operations to expand the training dataset, and the 5-fold cross-validation led to an accuracy of 98.61%. To evaluate the CNNs generalisation ability, we applied a cross-dataset validation technique. The obtained accuracies using cross-dataset experiments on three datasets were 97.04, 82.46 and 70.24%, which overcome the accuracies obtained by current state-of-the-art methods. We conclude that using the most common and deepest CNNs may not be the best choice for applications where the images to be classified differ from those used in pre-training. Additionally, the adopted cross-dataset validation approach proved to be an excellent choice to evaluate the generalisation capability of a model, as it considers the model performance on unseen data, which is paramount for CAD systems.

Recent computational methods for white blood cell nuclei segmentation: A comparative study.

BACKGROUND AND OBJECTIVE: Leukaemia is a disease found worldwide; it is a type of cancer that originates in the bone marrow and is characterised by an abnormal proliferation of white blood cells (leukocytes). In order to correctly identify this abnormality, haematologists examine blood smears from patients. A diagnosis obtained by this method may be influenced by factors such as the experience and level of fatigue of the haematologist, resulting in non-standard reports and even errors. In the literature, several methods have been proposed that involve algorithms to diagnose this disease. However, no reviews or surveys have been conducted. This paper therefore presents an empirical investigation of computational methods focusing on the segmentation of leukocytes. METHODS: In our study, 15 segmentation methods were evaluated using five public image databases: ALL-IDB2, BloodSeg, Leukocytes, JTSC Database and CellaVision. Following the standard methodology for literature evaluation, we conducted a pixel-level segmentation evaluation by comparing the segmented image with its corresponding ground truth. In order to identify the strengths and weaknesses of these methods, we performed an evaluation using six evaluation metrics: accuracy, specificity, precision, recall, kappa, Dice, and true positive rate. RESULTS: The segmentation algorithms performed significantly differently for different image databases, and for each database, a different algorithm achieved the best results. Moreover, the two best methods achieved average accuracy values higher than 97%, with an excellent kappa index. Also, the average Dice index indicated that the similarity between the segmented leukocyte and its ground truth was higher than 0.85 for these two methods This result confirms the high level of similarity between these images but does not guarantee that a method has segmented all leukocyte nuclei. We also found that the method that performed best segmented only 58.44% of all leukocytes. CONCLUSIONS: Of the techniques used to segment leukocytes, we note that clustering algorithms, the Otsu threshold, simple arithmetic operations and region growing are the approaches most widely used for this purpose. However, these computational methods have not yet overcome all the challenges posed by this problem.