Prediction of Osteoporosis Through Deep Learning Algorithms on Panoramic Radiographs

Objective: The aim of the present study was to predict osteoporosis on panoramic radiographs of women over 50 years of age through deep learning algorithms.Method: Panoramic radiographs of 744 female patients over 50 years of age were labeled as C1, C2, and C3 depending on mandibular cortical index (MCI). According to this index; C1: presence of a smooth and sharp mandibular cortex (normal); C2: resorption cavities at endosteal margin and 1 to 3-layer stratification (osteopenia); C3: completely porotic cortex (osteoporosis). The data of the present study were reviewed in different categories including C1-C2-C3, C1-C2, C1-C3 and C1-(C2+C3) as two-class and three-class prediction. The data were separated as 20% random test data; and the remaining data were used for training and validation with 5-fold cross-validation. AlexNET, GoogleNET, ResNET-50, SqueezeNET, and ShuffleNET deep learning models are trained through the transfer learning method. The results were evaluated by performance criteria including accuracy, sensitivity, specificity, F1-score, AUC and training duration. Findings: The dataset C1-C2-C3 has an accuracy rate of 81.14% with AlexNET; the dataset C1-C2 has an accuracy rate of 88.94% with GoogleNET; the dataset C1-C3 has an accuracy rate of 98.56% with AlexNET; and the dataset C1-(C2+C3) has an accuracy rate of 92.79% with GoogleNET. Conclusion: The highest accuracy was obtained in differentiation of C3 and C1 where osseous structure characteristics change significantly. Since the C2 score represent the intermediate stage (osteopenia), structural characteristics of the bone present behaviors closer to C1 and C3 scores. Therefore, the data set including the C2 score provided relatively lower accuracy results.

OViTAD: Optimized Vision Transformer to Predict Various Stages of Alzheimer's Disease Using Resting-State fMRI and Structural MRI Data

Advances in applied machine learning techniques to neuroimaging have encouraged scientists to implement models to early diagnose brain disorders such as Alzheimer's Disease. Predicting various stages of Alzheimer's disease is challenging; however, existing deep learning complex techniques could perform such a prediction. Therefore, using novel architectures with less complexity but efficient pattern extraction capabilities such as transformers has been of interest to neuroscientists. This study introduced an optimized vision transformer architecture to predict the aging effect in healthy adults (>75 years), mild cognitive impairment, and Alzheimer's' brains within the same age group using resting-state functional and anatomical magnetic resonance imaging data. Our optimized architecture known as OViTAD, which is currently the sole vision transformer-based end-to-end pipeline, outperformed the existing transformer models and most state-of-the-art solutions with F1-scores of 97%±0.0 and 0.9955%±0.0039 achieved from the testing sets for the two modalities in the triple-class prediction experiments where the number of trainable parameters decreased by 30% compared to a vanilla transformer. To ensure the robustness and reproducibility of our optimized vision transformer, we repeated the modeling process three times for all the experiments and reported the averaged evaluation metrics. Furthermore, we implemented a visualization technique to illustrate the effect of global attention on brain images. Also, we exhaustively implemented models to explore the impact of combining healthy brains with two other groups in the two modalities. This study could open a new avenue of adopting and optimizing vision transformers for neuroimaging applications, especially for Alzheimer's Disease prediction.

Drug Therapeutic-Use Class Prediction and Repurposing Using Graph Convolutional Networks

An important stage in the process of discovering new drugs is when candidate molecules are tested of their efficacy. It is reported that testing drug efficacy empirically costs billions of dollars in the drug discovery pipeline. As a mechanism of expediting this process, researchers have resorted to using computational methods to predict the action of molecules in silico. Here, we present a way of predicting the therapeutic-use class of drugs from chemical structures only using graph convolutional networks. In comparison with existing methods which use fingerprints or images as training samples, our approach has yielded better results in all metrics under consideration. In particular, validation accuracy increased from 83–88% to 86–90% for single label tasks. Similarly, the model achieved an accuracy of over 88% on new test data. Finally, our multi-label classification model made new predictions which indicated that some of the drugs could have other therapeutic uses other than those indicated in the dataset. We performed a literature-based evaluation of these predictions and found evidence that validates them. This renders the model a potential tool to be used in search of drugs that are candidates for repurposing.

A PROPOSED RECOGNITION SYSTEM FOR ALZHEIMER’S DISEASE BASED ON DEEP LEARNING AND OPTIMIZATION ALGORITHMS

Alzheimer's disease (AD) is an advanced and incurable neurodegenerative disease that causes progressive impairment of memory and cognitive functions due to the deterioration of brain cells. Early diagnosis is substantial to avoid permanent memory loss and develop treatments that will be subtracted in the future. Deep learning (DL) is a vital technique for medical imaging systems for AD diagnostics. The problem is multi-class classification seeking high accuracy. DL models have shown strong performance accuracy for multi-class prediction. In this paper, a proposed DL architecture is created to classify magnetic resonance imaging (MRI) to predict different stages of AD-based pre-trained Convolutional Neural Network (CNN) models and optimization algorithms. The proposed model architecture attempts to find the optimal subset of features to improve classification accuracy and reduce classification time. The pre-trained DL models, ResNet-101 and DenseNet-201, are utilized to extract features based on the last layer, and the Rival Genetic algorithm (RGA) and Pbest-Guide Binary Particle Swarm Optimization (PBPSO) are applied to select the optimal features. Then, the DL features and selected features are passed separately through created classifier for classification. The results are compared and analyzed by accuracy, performance metrics, and execution time. Experimental results showed that the most efficient accuracies were obtained by PBPSO selected features which reached 87.3% and 94.8% accuracy with less time of 46.7 sec, 32.7 sec for features based ResNet-101 and DenseNet-201, receptively.

Preoperative Differentiation of Pancreatic Cystic Neoplasm Subtypes on CT-Based Radiomics: SCN, MCN And IPMN

Background:The serous cystic neoplasm (SCN), mucinous cystic neoplasm (MCN), intraductal papillary mucinous neoplasm (IPMN) comprise the large proportion of pancreatic cystic neoplasm (PCN). The appropriate clinical management of MCN and IPMN isextremely essential to improve the 5-years survival rate for the early detection of pancreatic cancer. However, the differential diagnosis of patients with PCN before the treatment is still a tough challenge for all surgeons. Therefore, a reliable diagnosis tool is urgently required to be established for the improvement of precision diagnostics.Method:Between February 2016 and December 2020, 143 consecutive patients with PCN who were confirmed by postoperative pathology were retrospectively included in the study cohort, randomized into development and test cohort at the ratio of 7:3. The predictors of preoperative clinical-radiologic paraments were evaluated by the use of univariate and multivariable logistic regression analysis. A total of 1218 radiomics features were computationally extracted from the enhanced computed tomography (CT) of tumor region and a radiomics signature was established by the random forest algorithm. In the development cohort, the multi-class and binary-class radiomics models integrating preoperative variables and radiomics features were constructed to distinguish between the three types of PCN. The independent internal test cohort was applied to validate the classification models.Result:All preoperative prediction models were built by integrating the radiomics signature with thirteen diagnosis-related radiomics features and three important clinical-radiologic parameters of age, sex and tumor diameter. The multi-class prediction model presented an overall accuracy of 0.804 in the development cohort and 0.707 in the test cohort. The binary-class prediction models displayed the higher overall accuracy of 0.853, 0.866, 0.928 in the development dataset and 0.750, 0.839, 0.889 in the test dataset. In the test cohort, the binary-class radiomics models showed better predictive performances (AUC = 0.914, 0.863 ,0.926) than the multi-class radiomics model (AUC = 0.850), with a large net benefit in the decisive curve analysis. The radiomics-based nomogram provided the correct predicted probability for the diagnosis of PCN.Conclusion: The proposed radiomics models with clinical-radiologic parameters and radiomics features helped predict the accurate diagnosis among SCN, MCN, and IPMN to advance personalized medicine.

A DNA Methylation-Based Gene Signature Can Predict Triple-Negative Breast Cancer Diagnosis

Triple-negative breast cancer (TNBC) is the most aggressive breast cancer (BC) subtype and lacks targeted treatment. It is diagnosed by the absence of immunohistochemical expression of several biomarkers, but this method still displays some interlaboratory variability. DNA methylome aberrations are common in BC, thereby methylation profiling could provide the identification of accurate TNBC diagnosis biomarkers. Here, we generated a signature of differentially methylated probes with class prediction ability between 5 non-neoplastic breast and 7 TNBC tissues (error rate = 0.083). The robustness of this signature was corroborated in larger cohorts of additional 58 non-neoplastic breast, 93 TNBC, and 150 BC samples from the Gene Expression Omnibus repository, where it yielded an error rate of 0.006. Furthermore, we validated by pyrosequencing the hypomethylation of three out of 34 selected probes (FLJ43663, PBX Homeobox 1 (PBX1), and RAS P21 protein activator 3 (RASA3) in 51 TNBC, even at early stages of the disease. Finally, we found significantly lower methylation levels of FLJ43663 in cell free-DNA from the plasma of six TNBC patients than in 15 healthy donors. In conclusion, we report a novel DNA methylation signature with potential predictive value for TNBC diagnosis.

Deep Learning Features Encode Interpretable Morphologies within Histological Images

Convolutional neural networks (CNNs) are revolutionizing digital pathology by enabling machine learning-based classification of a variety of phenotypes from hematoxylin and eosin (H&E) whole slide images (WSIs), but the interpretation of CNNs remains difficult. Most studies have considered interpretability in a post hoc fashion, e.g. by presenting example regions with strongly predicted class labels. However, such an approach does not explain the biological features that contribute to correct predictions. To address this problem, here we investigate the interpretability of H&E-derived CNN features (the feature weights in the final layer of a transfer-learning-based architecture), which we show can be construed as abstract morphological genes (“mones”) with strong independent associations to biological phenotypes. We observe that many mones are specific to individual cancer types, while others are found in multiple cancers especially from related tissue types. We also observe that mone-mone correlations are strong and robustly preserved across related cancers. Importantly, linear mone-based classifiers can very accurately separate 38 distinct classes (19 tumor types and their adjacent normals, AUC=97.1%±2.8% for each class prediction), and linear classifiers are also highly effective for universal tumor detection (AUC=99.2%±0.12%). This linearity provides evidence that individual mones or correlated mone clusters may be associated with interpretable histopathological features or other patient characteristics. In particular, the statistical similarity of mones to gene expression values allows integrative mone analysis via expression-based bioinformatics approaches. We observe strong correlations between individual mones and individual gene expression values, notably mones associated with collagen gene expression in ovarian cancer. Mone-expression comparisons also indicate that immunoglobulin expression can be identified using mones in colon adenocarcinoma and that immune activity can be identified across multiple cancer types, and we verify these findings by expert histopathological review. Our work demonstrates that mones provide a morphological H&E decomposition that can be effectively associated with diverse phenotypes, analogous to the interpretability of transcription via gene expression values.