Unveiling Thymoma Typing Through Hyperspectral Imaging and Deep Learning.

Thymoma, a rare tumor from thymic epithelial cells, presents diagnostic challenges because of the subjective nature of traditional methods, leading to high false-negative rates and long diagnosis times. This study introduces a thymoma classification technique that integrates hyperspectral imaging with deep learning. We initially capture pathological slice images of thymoma using a hyperspectral camera and delineate regions of interest to extract spectral data. This data undergoes reflectance calibration and noise reduction. Subsequently, we transform the spectral data into two-dimensional images via the Gramian Angular Field (GAF) method. A variant residual network is then utilized to extract features and classify these images. Our results demonstrate that this model significantly enhances classification accuracy and efficiency, achieving an average accuracy of 95%. The method proves highly effective in automated thymoma diagnosis, optimizing data utilization, and feature representation learning.

Harnessing Deep Learning for Accurate Pathological Assessment of Brain Tumor Cell Types.

Primary diffuse central nervous system large B-cell lymphoma (CNS-pDLBCL) and high-grade glioma (HGG) often present similarly, clinically and on imaging, making differentiation challenging. This similarity can complicate pathologists' diagnostic efforts, yet accurately distinguishing between these conditions is crucial for guiding treatment decisions. This study leverages a deep learning model to classify brain tumor pathology images, addressing the common issue of limited medical imaging data. Instead of training a convolutional neural network (CNN) from scratch, we employ a pre-trained network for extracting deep features, which are then used by a support vector machine (SVM) for classification. Our evaluation shows that the Resnet50 (TL + SVM) model achieves a 97.4% accuracy, based on tenfold cross-validation on the test set. These results highlight the synergy between deep learning and traditional diagnostics, potentially setting a new standard for accuracy and efficiency in the pathological diagnosis of brain tumors.

Identification of gastric cancer types based on hyperspectral imaging technology.

Gastric cancer is becoming the second biggest cause of death from cancer. Treatment and prognosis of different types of gastric cancer vary greatly. However, the routine pathological examination is limited to the tissue level and is easily affected by subjective factors. In our study, we examined gastric mucosal samples from 50 normal tissue and 90 cancer tissues. Hyperspectral imaging technology was used to obtain spectral information. A two-classification model for normal tissue and cancer tissue identification and a four-classification model for cancer type identification are constructed based on the improved deep residual network (IDRN). The accuracy of the two-classification model and four-classification model are 0.947 and 0.965. Hyperspectral imaging technology was used to extract molecular information to realize real-time diagnosis and accurate typing. The results show that hyperspectral imaging technique has good effect on diagnosis and type differentiation of gastric cancer, which is expected to be used in auxiliary diagnosis and treatment.

Graded diagnosis of Helicobacter pylori infection using hyperspectral images of gastric juice.

Helicobacter pylori is a potential underlying cause of many diseases. Although the Carbon 13 breath test is considered the gold standard for detection, it is high cost and low public accessibility in certain areas limit its widespread use. In this study, we sought to use machine learning and deep learning algorithm models to classify and diagnose H. pylori infection status. We used hyperspectral imaging system to gather gastric juice images and then retrieved spectral feature information between 400 and 1000 nm. Two different data processing methods were employed, resulting in the establishment of one-dimensional (1D) and two-dimensional (2D) datasets. In the binary classification task, the random forest model achieved a prediction accuracy of 83.27% when learning features from 1D data, with a specificity of 84.56% and a sensitivity of 92.31%. In the ternary classification task, the ResNet model learned from 2D data and achieved a classification accuracy of 91.48%.

Accurate classification of glomerular diseases by hyperspectral imaging and transformer.

BACKGROUND AND OBJECTIVE: In renal disease research, precise glomerular disease diagnosis is crucial for treatment and prognosis. Currently reliant on invasive biopsies, this method bears risks and pathologist-dependent variability, yielding inconsistent results. There is a pressing need for innovative diagnostic tools that enhance traditional methods, streamline processes, and ensure accurate and consistent disease detection. METHODS: In this study, we present an innovative Convolutional Neural Networks-Vision Transformer (CVT) model leveraging Transformer technology to refine glomerular disease diagnosis by fusing spectral and spatial data, surpassing traditional diagnostic limitations. Using interval sampling, preprocessing, and wavelength optimization, we also introduced the Gramian Angular Field (GAF) method for a unified representation of spectral and spatial characteristics. RESULTS: We captured hyperspectral images ranging from 385.18 nm to 1009.47 nm and employed various methods to extract sample features. Initial models based solely on spectral features achieved a accuracy of 85.24 %. However, the CVT model significantly outperformed these, achieving an average accuracy of 94 %. This demonstrates the model's superior capability in utilizing sample data and learning joint feature representations. CONCLUSIONS: The CVT model not only breaks through the limitations of existing diagnostic techniques but also showcases the vast potential of non-invasive, high-precision diagnostic technology in supporting the classification and prognosis of complex glomerular diseases. This innovative approach could significantly impact future diagnostic strategies in renal disease research. CONCISE ABSTRACT: This study introduces a transformative hyperspectral image classification model leveraging a Transformer to significantly improve glomerular disease diagnosis accuracy by synergizing spectral and spatial data, surpassing conventional methods. Through a rigorous comparative analysis, it was determined that while spectral features alone reached a peak accuracy of 85.24 %, the novel Convolutional Neural Network-Transformer (CVT) model's integration of spatial-spectral features via the Gramian Angular Field (GAF) method markedly enhanced diagnostic precision, achieving an average accuracy of 94 %. This methodological innovation not only overcomes traditional diagnostic limitations but also underscores the potential of non-invasive, high-precision technologies in advancing the classification and prognosis of complex renal diseases, setting a new benchmark in the field.

The Contrastive Network With Convolution and Self-Attention Mechanisms for Unsupervised Cell Segmentation.

Deep learning for cell instance segmentation is a significant research direction in biomedical image analysis. The traditional supervised learning methods rely on pixel-wise annotation of object images to train the models, which is often accompanied by time-consuming and labor-intensive. Various modified segmentation methods, based on weakly supervised or semi-supervised learning, have been proposed to recognize cell regions by only using rough annotations of cell positions. However, it is still hard to achieve the fully unsupervised in most approaches that the utilization of few annotations for training is still inevitable. In this article, we propose an end-to-end unsupervised model that can segment individual cell regions on hematoxylin and eosin (H&E) stained slides without any annotation. Compared with weakly or semi-supervised methods, the input of our model is in the form of raw data without any identifiers and there is no need to generate pseudo-labelling during training. We demonstrated that the performance of our model is satisfactory and also has a great generalization ability on various validation sets compared with supervised models. The ablation experiment shows that our backbone has superior performance in capturing object edge and context information than pure CNN or transformer under our unsupervised method.

Construction of artificial intelligence non-invasive diagnosis model for common glomerular diseases based on hyperspectral and urine analysis.

OBJECTIVE: To develop a non-invasive fluid biopsy assisted diagnosis model for glomerular diseases based on hyperspectral, so as to solve the problem of poor compliance of patients with invasive examination and improve the early diagnosis rate of glomerular diseases. METHODS: A total of 65 urine samples from patients who underwent renal biopsy from November 2020 to January 2022 in Qianfoshan Hospital of Shandong Province were collected.By simultaneously capturing spectral information of the above urine samples in the 400-1000 nm range, more obvious differences were found in the spectra of urine from patients with glomerular diseases between 650 nm and 680 nm. We obtained the original hyperspectral images in this wavelength range through digital scanning, and sampled pixel points at intervals on the original images. The two-dimensional digital image generated from each pixel point served as a member of the subsequent training and test sets. . After manually labeling the images according to different biopsy pathological types, they were randomly divided into training set (n = 58,800) and test set (n = 25,200). The training set was used for training learning and parameter iteration of artificial intelligence non-invasive liquid diagnosis model, and the test set for model recognition and interpretation. The evaluation indexes such as accuracy, sensitivity and specificity were calculated to evaluate the performance of the diagnosis model. RESULTS: The model has an accuracy rate of 96% for early diagnosis of four glomerular diseases. CONCLUSION: The auxiliary diagnosis model system has high accuracy. It is expected to be used as a non-invasive diagnostic method for glomerular diseases in clinic.

Research for accurate auxiliary diagnosis of lung cancer based on intracellular fluorescent fingerprint information.

The distinctions in pathological types and genetic subtypes of lung cancer have a direct impact on the choice of treatment choices and clinical prognosis in clinical practice. This study used pathological histological sections of surgically removed or biopsied tumor tissue from 36 patients. Based on a small sample size, millions of spectral data points were extracted to investigate the feasibility of employing intracellular fluorescent fingerprint information to diagnose the pathological types and mutational status of lung cancer. The intracellular fluorescent fingerprint information revealed the EGFR gene mutation characteristics in lung cancer, and the area under the curve (AUC) value for the optimal model was 0.98. For the classification of lung cancer pathological types, the macro average AUC value for the ensemble-learning model was 0.97. Our research contributes new idea for pathological diagnosis of lung cancer and offers a quick, easy, and accurate auxiliary diagnostic approach.

Combining hyperspectral imaging techniques with deep learning to aid in early pathological diagnosis of melanoma.

BACKGROUND: Cutaneous melanoma, an exceedingly aggressive form of skin cancer, holds the top rank in both malignancy and mortality among skin cancers. In early stages, distinguishing malignant melanomas from benign pigmented nevi pathologically becomes a significant challenge due to their indistinguishable traits. Traditional skin histological examination techniques, largely reliant on light microscopic imagery, offer constrained information and yield low-contrast results, underscoring the necessity for swift and effective early diagnostic methodologies. As a non-contact, non-ionizing, and label-free imaging tool, hyperspectral imaging offers potential in assisting pathologists with identification procedures sans contrast agents. METHODS: This investigation leverages hyperspectral cameras to ascertain the optical properties and to capture the spectral features of malignant melanoma and pigmented nevus tissues, intending to facilitate early pathological diagnostic applications. We further enhance the diagnostic process by integrating transfer learning with deep convolutional networks to classify melanomas and pigmented nevi in hyperspectral pathology images. The study encompasses pathological sections from 50 melanoma and 50 pigmented nevus patients. To accurately represent the spectral variances between different tissues, we employed reflectance calibration, highlighting that the most distinctive spectral differences emerged within the 500-675 nm band range. RESULTS: The classification accuracy of pigmented tumors and pigmented nevi was 89% for one-dimensional sample data and 98% for two-dimensional sample data. CONCLUSIONS: Our findings have the potential to expedite pathological diagnoses, enhance diagnostic precision, and offer novel research perspectives in differentiating melanoma and nevus.