In-house Dataset Research Articles

SPINEPS—automatic whole spine segmentation of T2-weighted MR images using a two-phase approach to multi-class semantic and instance segmentation

Abstract Objectives Introducing SPINEPS, a deep learning method for semantic and instance segmentation of 14 spinal structures (ten vertebra substructures, intervertebral discs, spinal cord, spinal canal, and sacrum) in whole-body sagittal T2-weighted turbo spin echo images. Material and methods This local ethics committee-approved study utilized a public dataset (train/test 179/39 subjects, 137 female), a German National Cohort (NAKO) subset (train/test 1412/65 subjects, mean age 53, 694 female), and an in-house dataset (test 10 subjects, mean age 70, 5 female). SPINEPS is a semantic segmentation model, followed by a sliding window approach utilizing a second model to create instance masks from the semantic ones. Segmentation evaluation metrics included the Dice score and average symmetrical surface distance (ASSD). Statistical significance was assessed using the Wilcoxon signed-rank test. Results On the public dataset, SPINEPS outperformed a nnUNet baseline on every structure and metric (e.g., an average over vertebra instances: dice 0.933 vs 0.911, p < 0.001, ASSD 0.21 vs 0.435, p < 0.001). SPINEPS trained on automated annotations of the NAKO achieves an average global Dice score of 0.918 on the combined NAKO and in-house test split. Adding the training data from the public dataset outperforms this (average instance-wise Dice score over the vertebra substructures 0.803 vs 0.778, average global Dice score 0.931 vs 0.918). Conclusion SPINEPS offers segmentation of 14 spinal structures in T2w sagittal images. It provides a semantic mask and an instance mask separating the vertebrae and intervertebral discs. This is the first publicly available algorithm to enable this segmentation. Key Points QuestionNo publicly available automatic approach can yield semantic and instance segmentation masks for the whole spine (including posterior elements) in T2-weighted sagittal TSE images. FindingsSegmenting semantically first and then instance-wise outperforms a baseline trained directly on instance segmentation. The developed model produces high-resolution MRI segmentations for the whole spine. Clinical relevanceThis study introduces an automatic approach to whole spine segmentation, including posterior elements, in arbitrary fields of view T2w sagittal MR images, enabling easy biomarker extraction, automatic localization of pathologies and degenerative diseases, and quantifying analyses as downstream research.

Transformer-Integrated Hybrid Convolutional Neural Network for Dose Prediction in Nasopharyngeal Carcinoma Radiotherapy.

Radiotherapy is recognized as the major treatment of nasopharyngeal carcinoma. Rapid and accurate dose prediction can improve the efficiency of the treatment planning process and the quality of radiotherapy plans. Currently, deep learning-based methods have been widely applied to dose prediction for radiotherapy treatment planning. However, it is important to note that existing models based on Convolutional Neural Networks (CNN) often overlook long-distance information. Although some studies try to use Transformer to solve the problem, it lacks the ability of CNN to process the spatial information inherent in images. Therefore, we propose a novel CNN and Transformer hybrid dose prediction model. To enhance the transmission ability of features between CNN and Transformer, we design a hierarchical dense recurrent encoder with a channel attention mechanism. Additionally, we propose a progressive decoder that preserves richer texture information through layer-wise reconstruction of high-dimensional feature maps. The proposed model also introduces object-driven skip connections, which facilitate the flow of information between the encoder and decoder. Experiments are conducted on in-house datasets, and the results show that the proposed model is superior to baseline methods in most dosimetric criteria. In addition, the image analysis metrics including PSNR, SSIM, and NRMSE demonstrate that the proposed model is consistent with ground truth and produces promising visual effects compared to other advanced methods. The proposed method could be taken as a powerful clinical guidance tool for physicists, significantly enhancing the efficiency of radiotherapy planning. The source code is available at https://github.com/CDUTJ102/THCN-Net .

Machine learning-derived peripheral blood transcriptomic biomarkers for early lung cancer diagnosis: Unveiling tumor-immune interaction mechanisms.

Lung cancer continues to be the leading cause of cancer-related mortality worldwide. Early detection and a comprehensive understanding of tumor-immune interactions are crucial for improving patient outcomes. This study aimed to develop a novel biomarker panel utilizing peripheral blood transcriptomics and machine learning algorithms for early lung cancer diagnosis, while simultaneously providing insights into tumor-immune crosstalk mechanisms. Leveraging a training cohort (GSE135304), we employed multiple machine learning algorithms to formulate a Lung Cancer Diagnostic Score (LCDS) based on peripheral blood transcriptomic features. The LCDS model's performance was evaluated using the area under the receiver operating characteristic (ROC) curve (AUC) in multiple validation cohorts (GSE42834, GSE157086, and an in-house dataset). Peripheral blood samples were obtained from 20 lung cancer patients and 10 healthy control subjects, representing an in-house cohort recruited at the Sixth People's Hospital of Chengdu. We employed advanced bioinformatics techniques to explore tumor-immune interactions through comprehensive immune infiltration and pathway enrichment analyses. Initial screening identified 844 differentially expressed genes, which were subsequently refined to 87 genes using the Boruta feature selection algorithm. The random forest (RF) algorithm demonstrated the highest accuracy in constructing the LCDS model, yielding a mean AUC of 0.938. Lower LCDS values were significantly associated with elevated immune scores and increased CD4+ and CD8+ T-cell infiltration, indicative of enhanced antitumor-immune responses. Higher LCDS scores correlated with activation of hypoxia, peroxisome proliferator-activated receptor (PPAR), and Toll-like receptor (TLR) signaling pathways, as well as reduced DNA damage repair pathway scores. Our study presents a novel, machine learning-derived peripheral blood transcriptomic biomarker panel with potential applications in early lung cancer diagnosis. The LCDS model not only demonstrates high accuracy in distinguishing lung cancer patients from healthy individuals but also offers valuable insights into tumor-immune interactions and underlying cancer biology. This approach may facilitate early lung cancer detection and contribute to a deeper understanding of the molecular and cellular mechanisms underlying tumor-immune crosstalk. Furthermore, our findings on the relationship between LCDS and immune infiltration patterns may have implications for future research on therapeutic strategies targeting the immune system in lung cancer.

Investigating PPT2’s role in ovarian cancer prognosis and immunotherapy outcomes

Ovarian cancer (OC) remains the primary cause of mortality among gynecological malignancies, and the identification of reliable molecular biomarkers to prognosticate OC outcomes is yet to be achieved. The gene palmitoyl protein thioesterase 2 (PPT2), which has been sparsely studied in OC, was closely associated with metabolism. This study aimed to determine the association between PPT2 expression, prognosis, immune infiltration, and potential molecular mechanisms in OC. We obtained the RNA-seq and clinical data from The Cancer Genome Atlas (TCGA), The Genotype-Tissue Expression (GTEx) and Gene Expression Omnibus (GEO) databases, then Kaplan-Meier analysis, univariate Cox regression, multivariate Cox regression, nomogram, and calibration were conducted to assess and verify the role of PPT2. Gene set enrichment analysis (GSEA) was used to figure out the closely correlated pathways with PPT2. Overexpression experiment was performed to explore the function of PPT2. Our findings showed that PPT2 mRNA expression was apparent down-regulation in OC tissue compared to normal ovarian tissues in TCGA, GTEx datasets, and GEO datasets. This differential expression was also confirmed in our in-house datasets at both the mRNA and protein levels. Decreased PPT2 expression correlated with lower survival rates in TCGA, several GEO datasets, and our in-house datasets. Multivariate analysis revealed that PPT2 was an independent factor in predicting better outcomes for OC patients in TCGA and GEO. A negative correlation was revealed between immune infiltration and PPT2 expression through Single-sample GSEA (ssGSEA). Additionally, PPT2 was negatively correlated with an up-regulated immune score, stromal score, and estimate score, suggesting that patients with low PPT2 expression might benefit more from immunotherapy. Numerous chemical agents showed lower IC50 in patients with high PPT2 expression. In single-cell RNA sequencing (scRNA-seq) analysis of several OC datasets, we found PPT2 was mainly expressed in endothelial cells. Furthermore, we found that PPT2 inhibited OC cell proliferation in vitro. Our results demonstrated that PPT2 was considered a favorable prognostic biomarker for OC and may be vital in predicting response to immunotherapy and chemotherapy. Further research was needed to fully understand the relationship between PPT2 and immunotherapy efficacy in OC patients.

An anthropomorphic diagnosis system of pulmonary nodules using weak annotation-based deep learning

The accurate categorization of lung nodules in CT scans is an essential aspect in the prompt detection and diagnosis of lung cancer. The categorization of grade and texture for nodules is particularly significant since it can aid radiologists and clinicians to make better-informed decisions concerning the management of nodules. However, currently existing nodule classification techniques have a singular function of nodule classification and rely on an extensive amount of high-quality annotation data, which does not meet the requirements of clinical practice. To address this issue, we develop an anthropomorphic diagnosis system of pulmonary nodules (PN) based on deep learning (DL) that is trained by weak annotation data and has comparable performance to full-annotation based diagnosis systems. The proposed system uses DL models to classify PNs (benign vs. malignant) with weak annotations, which eliminates the need for time-consuming and labor-intensive manual annotations of PNs. Moreover, the PN classification networks, augmented with handcrafted shape features acquired through the ball-scale transform technique, demonstrate capability to differentiate PNs with diverse labels, including pure ground-glass opacities, part-solid nodules, and solid nodules. Through 5-fold cross-validation on two datasets, the system achieved the following results: (1) an Area Under Curve (AUC) of 0.938 for PN localization and an AUC of 0.912 for PN differential diagnosis on the LIDC-IDRI dataset of 814 testing cases, (2) an AUC of 0.943 for PN localization and an AUC of 0.815 for PN differential diagnosis on the in-house dataset of 822 testing cases. In summary, our system demonstrates efficient localization and differential diagnosis of PNs in a resource limited environment, and thus could be translated into clinical use in the future.

Optimizing Catheter Verification

Purpose Accurate detection of central venous catheter (CVC) misplacement is crucial for patient safety and effective treatment. Existing artificial intelligence (AI) often grapple with the limitations of label inaccuracies and output interpretations that lack clinician-friendly comprehensibility. This study aims to introduce an approach that employs segmentation of support material and anatomy to enhance the precision and comprehensibility of CVC misplacement detection. Materials and Methods The study utilized 2 datasets: the publicly accessible RANZCR CLiP dataset and a bespoke in-house dataset of 1006 annotated supine chest x-rays. Three deep learning models were trained: a classification network, a segmentation network, and a combination of both. These models were evaluated using receiver operating characteristic analysis, area under the curve, DICE similarity coefficient, and Hausdorff distance. Results The combined model demonstrated superior performance with an area under the curve of 0.99 for correctly positioned CVCs and 0.95 for misplacements. The model maintained high efficacy even with reduced training data from the local dataset. Sensitivity and specificity rates were high, and the model effectively managed the segmentation and classification tasks, even in images with multiple CVCs and other support materials. Conclusions This study illustrates the potential of AI-based models in accurately and reliably determining CVC placement in chest x-rays. The proposed method shows high accuracy and offers improved interpretability, important for clinical decision-making. The findings also highlight the importance of dataset quality and diversity in training AI models for medical image analysis.

Multi-contrast image super-resolution with deformable attention and neighborhood-based feature aggregation (DANCE): Applications in anatomic and metabolic MRI

Multi-contrast magnetic resonance imaging (MRI) reflects information about human tissues from different perspectives and has wide clinical applications. By utilizing the auxiliary information from reference images (Refs) in the easy-to-obtain modality, multi-contrast MRI super-resolution (SR) methods can synthesize high-resolution (HR) images from their low-resolution (LR) counterparts in the hard-to-obtain modality. In this study, we systematically discussed the potential impacts caused by cross-modal misalignments between LRs and Refs and, based on this discussion, proposed a novel deep-learning-based method with Deformable Attention and Neighborhood-based feature aggregation to be Computationally Efficient (DANCE) and insensitive to misalignments. Our method has been evaluated in two public MRI datasets, i.e., IXI and FastMRI, and an in-house MR metabolic imaging dataset with amide proton transfer weighted (APTW) images. Experimental results reveal that our method consistently outperforms baselines in various scenarios, with significant superiority observed in the misaligned group of IXI dataset and the prospective study of the clinical dataset. The robustness study proves that our method is insensitive to misalignments, maintaining an average PSNR of 30.67 dB when faced with a maximum range of ±9°and ±9 pixels of rotation and translation on Refs. Given our method’s desirable comprehensive performance, good robustness, and moderate computational complexity, it possesses substantial potential for clinical applications.

A 3D approach to understanding heterogeneity in early developing autisms.

Phenotypic heterogeneity in early language, intellectual, motor, and adaptive functioning (LIMA) features are amongst the most striking features that distinguish different types of autistic individuals. Yet the current diagnostic criteria uses a single label of autism and implicitly emphasizes what individuals have in common as core social-communicative and restricted repetitive behavior difficulties. Subtype labels based on the non-core LIMA features may help to more meaningfully distinguish types of autisms with differing developmental paths and differential underlying biology. Unsupervised data-driven subtypes were identified using stability-based relative clustering validation on publicly available Mullen Scales of Early Learning (MSEL) and Vineland Adaptive Behavior Scales (VABS) data (n = 615; age = 24-68months) from the National Institute of Mental Health Data Archive (NDA). Differential developmental trajectories between subtypes were tested on longitudinal data from NDA and from an independent in-house dataset from UCSD. A subset of the UCSD dataset was also tested for subtype differences in functional and structural neuroimaging phenotypes and relationships with blood gene expression. The current subtyping model was also compared to early language outcome subtypes derived from past work. Two autism subtypes can be identified based on early phenotypic LIMA features. These data-driven subtypes are robust in the population and can be identified in independent data with 98% accuracy. The subtypes can be described as Type I versus Type II autisms differentiated by relatively high versus low scores on LIMA features. These two types of autisms are also distinguished by different developmental trajectories over the first decade of life. Finally, these two types of autisms reveal striking differences in functional and structural neuroimaging phenotypes and their relationships with gene expression and may highlight unique biological mechanisms. Sample sizes for the neuroimaging and gene expression dataset are relatively small and require further independent replication. The current work is also limited to subtyping based on MSEL and VABS phenotypic measures. This work emphasizes the potential importance of stratifying autism by a Type I versus Type II distinction focused on LIMA features and which may be of high prognostic and biological significance.

Multi-View Integrative Approach For Imputing Short-Chain Fatty Acids and Identifying Key factors predicting Blood SCFA.

Short-chain fatty acids (SCFAs) are the main metabolites produced by bacterial fermentation of dietary fiber within gastrointestinal tract. SCFAs produced by gut microbiotas (GMs) are absorbed by host, reach bloodstream, and are distributed to different organs, thus influencing host physiology. However, due to the limited budget or the poor sensitivity of instruments, most studies on GMs have incomplete blood SCFA data, limiting our understanding of the metabolic processes within the host. To address this gap, we developed an innovative multi-task multi-view integrative approach (M2AE, Multi-task Multi-View Attentive Encoders), to impute blood SCFA levels using gut metagenomic sequencing (MGS) data, while taking into account the intricate interplay among the gut microbiome, dietary features, and host characteristics, as well as the nuanced nature of SCFA dynamics within the body. Here, each view represents a distinct type of data input (i.e., gut microbiome compositions, dietary features, or host characteristics). Our method jointly explores both view-specific representations and cross-view correlations for effective predictions of SCFAs. We applied M2AE to two in-house datasets, which both include MGS and blood SCFAs profiles, host characteristics, and dietary features from 964 subjects and 171 subjects, respectively. Results from both of two datasets demonstrated that M2AE outperforms traditional regression-based and neural-network based approaches in imputing blood SCFAs. Furthermore, a series of gut bacterial species (e.g., Bacteroides thetaiotaomicron and Clostridium asparagiforme), host characteristics (e.g., race, gender), as well as dietary features (e.g., intake of fruits, pickles) were shown to contribute greatly to imputation of blood SCFAs. These findings demonstrated that GMs, dietary features and host characteristics might contribute to the complex biological processes involved in blood SCFA productions. These might pave the way for a deeper and more nuanced comprehension of how these factors impact human health.

Developing Robust Human Liver Microsomal Stability Prediction Models: Leveraging Inter-Species Correlation with Rat Data.

Objectives: Pharmacokinetic issues were the leading cause of drug attrition, accounting for approximately 40% of all cases before the turn of the century. To this end, several high-throughput in vitro assays like microsomal stability have been developed to evaluate the pharmacokinetic profiles of compounds in the early stages of drug discovery. At NCATS, a single-point rat liver microsomal (RLM) stability assay is used as a Tier I assay, while human liver microsomal (HLM) stability is employed as a Tier II assay. We experimentally screened and collected data on over 30,000 compounds for RLM stability and over 7000 compounds for HLM stability. Although HLM stability screening provides valuable insights, the increasing number of hits generated, along with the time- and resource-intensive nature of the assay, highlights the need for alternative strategies. One promising approach is leveraging in silico models trained on these experimental datasets. Methods: We describe the development of an HLM stability prediction model using our in-house HLM stability dataset. Results: Employing both classical machine learning methods and advanced techniques, such as neural networks, we achieved model accuracies exceeding 80%. Moreover, we validated our model using external test sets and found that our models are comparable to some of the best models in literature. Additionally, the strong correlation observed between our RLM and HLM data was further reinforced by the fact that our HLM model performance improved when using RLM stability predictions as an input descriptor. Conclusions: The best model along with a subset of our dataset (PubChem AID: 1963597) has been made publicly accessible on the ADME@NCATS website for the benefit of the greater drug discovery community. To the best of our knowledge, it is the largest open-source model of its kind and the first to leverage cross-species data.

DSANet: Dual-path segmentation-guided attention network for radiotherapy dose prediction from CT images only

In radiation therapy, an ideal treatment plan always takes the treatment planner a couple of days to tweak repeatedly, owing to the complex dose distribution which requires the prescribed dose in the tumor area but significantly reduced radiation dose in the surrounding normal tissues. To facilitate this process, many deep learning-based studies have been devoted to the automatical prediction of dose distribution. However, besides the computed tomography (CT) image, these methods usually require extra segmentation masks such as tumor target and organs at risk (OARs) as prior knowledge, and acquiring such annotations are quite time-consuming. In this paper, we propose a Dual-path Segmentation-guided Attention Network (DSANet) to directly predict the dose distribution map solely from the CT images. Specifically, considering the high correlation between the tumor anatomical structure and the dose distribution, we develop a dual-path architecture which consists of 1) a primary dose prediction path aiming at generating accurate dose distribution maps, and 2) an auxiliary segmentation path applied to provide attention maps with anatomical information for the primary dose prediction task. Moreover, we design an attention map-guided feature decoupling (AMFD) module to enforce the primary network to focus not only on the high-dose distribution in the tumor area but also on the low-dose distribution in the surrounding regions. Experiments on two in-house datasets demonstrate the effectiveness and generalization of our method.

LesionMix data enhancement and entropy minimization for semi-supervised lesion segmentation of lung cancer

Determining the location and contour of the lesion is a crucial prerequisite for medical diagnosis, subsequent personalized treatment plan and prognostic prediction of lung cancer. Semi-supervised learning and data augmentation methods facilitate deep learning to be used in many fields of medical imaging. In this paper, we introduce a novel data enhancement technique called LesionMix. This method involves extracting and reusing lesions from a limited amount of labeled CT data, thereby enhancing the efficiency of utilizing those labeled data. Meanwhile, we propose a two-stage semi-supervised training strategy called Entropy Minimization LesionMix (EMLM). In the first stage, features containing lesion contour information are rapidly learned through LesionMix data augmentation. Entropy minimization strategy optimizes the model parameters to alleviate overfitting as much as possible in the second stage and improves prediction confidence. Our proposed method is validated on the public dataset LIDC-IDRI and the in-house dataset GDPHLUAD. Extensive experiments demonstrate that our method achieves promising performance and outperforms seven state-of-the-art semi-supervised models; Moreover, ablation experiments validate the effectiveness of various aspects of our approach.

Re-identification from histopathology images

In numerous studies, deep learning algorithms have proven their potential for the analysis of histopathology images, for example, for revealing the subtypes of tumors or the primary origin of metastases. These models require large datasets for training, which must be anonymized to prevent possible patient identity leaks. This study demonstrates that even relatively simple deep learning algorithms can re-identify patients in large histopathology datasets with substantial accuracy. In addition, we compared a comprehensive set of state-of-the-art whole slide image classifiers and feature extractors for the given task. We evaluated our algorithms on two TCIA datasets including lung squamous cell carcinoma (LSCC) and lung adenocarcinoma (LUAD). We also demonstrate the algorithm’s performance on an in-house dataset of meningioma tissue. We predicted the source patient of a slide with F1 scores of up to 80.1% and 77.19% on the LSCC and LUAD datasets, respectively, and with 77.09% on our meningioma dataset. Based on our findings, we formulated a risk assessment scheme to estimate the risk to the patient’s privacy prior to publication.

Semi-supervised ViT knowledge distillation network with style transfer normalization for colorectal liver metastases survival prediction

Colorectal liver metastases (CLM) affect almost half of all colon cancer patients and the response to systemic chemotherapy plays a crucial role in patient survival. While oncologists typically use tumor grading scores, such as tumor regression grade (TRG), to establish an accurate prognosis on patient outcomes, including overall survival (OS) and time-to-recurrence (TTR), these traditional methods have several limitations. They are subjective, time-consuming, and require extensive expertise, which limits their scalability and reliability. Additionally, existing approaches for prognosis prediction using machine learning mostly rely on radiological imaging data, but recently histological images have been shown to be relevant for survival predictions by allowing to fully capture the complex microenvironmental and cellular characteristics of the tumor. To address these limitations, we propose an end-to-end approach for automated prognosis prediction using histology slides stained with Hematoxylin and Eosin (H&E) and Hematoxylin Phloxine Saffron (HPS). We first employ a Generative Adversarial Network (GAN) for slide normalization to reduce staining variations and improve the overall quality of the images that are used as input to our prediction pipeline. We propose a semi-supervised model to perform tissue classification from sparse annotations, producing segmentation and feature maps. Specifically, we use an attention-based approach that weighs the importance of different slide regions in producing the final classification results. Finally, we exploit the extracted features for the metastatic nodules and surrounding tissue to train a prognosis model. In parallel, we train a vision Transformer model in a knowledge distillation framework to replicate and enhance the performance of the prognosis prediction. We evaluate our approach on an in-house clinical dataset of 258 CLM patients, achieving superior performance compared to other comparative models with a c-index of 0.804 (0.014) for OS and 0.735 (0.016) for TTR, as well as on two public datasets. The proposed approach achieves an accuracy of 86.9% to 90.3% in predicting TRG dichotomization. For the 3-class TRG classification task, the proposed approach yields an accuracy of 78.5% to 82.1%, outperforming the comparative methods. Our proposed pipeline can provide automated prognosis for pathologists and oncologists, and can greatly promote precision medicine progress in managing CLM patients.

A robust image segmentation and synthesis pipeline for histopathology

Significant diagnostic variability between and within observers persists in pathology, despite the fact that digital slide images provide the ability to measure and quantify features much more precisely compared to conventional methods. Automated and accurate segmentation of cancerous cell and tissue regions can streamline the diagnostic process, providing insights into the cancer progression, and helping experts decide on the most effective treatment. Here, we evaluate the performance of the proposed PathoSeg model, with an architecture comprising of a modified HRNet encoder and a UNet++ decoder integrated with a CBAM block to utilize attention mechanism for an improved segmentation capability. We demonstrate that PathoSeg outperforms the current state-of-the-art (SOTA) networks in both quantitative and qualitative assessment of instance and semantic segmentation. Notably, we leverage the use of synthetic data generated by PathopixGAN, which effectively addresses the data imbalance problem commonly encountered in histopathology datasets, further improving the performance of PathoSeg. It utilizes spatially adaptive normalization within a generative and discriminative mechanism to synthesize diverse histopathological environments dictated through semantic information passed through pixel-level annotated Ground Truth semantic masks.Besides, we contribute to the research community by providing an in-house dataset that includes semantically segmented masks for breast carcinoma tubules (BCT), micro/macrovesicular steatosis of the liver (MSL), and prostate carcinoma glands (PCG). In the first part of the dataset, we have a total of 14 whole slide images from 13 patients’ liver, with fat cell segmented masks, totaling 951 masks of size 512 × 512 pixels. In the second part, it includes 17 whole slide images from 13 patients with prostate carcinoma gland segmentation masks, amounting to 30,000 masks of size 512 × 512 pixels. In the third part, the dataset contains 51 whole slides from 36 patients, with breast carcinoma tubule masks totaling 30,000 masks of size 512 × 512 pixels. To ensure transparency and encourage further research, we will make this dataset publicly available for non-commercial and academic purposes. To facilitate reproducibility and encourage further research, we will also make our code and pre-trained models publicly available at https://github.com/DeepMIALab/PathoSeg.

Practical feature filter strategy to machine learning for small datasets in chemistry

Many potential use cases for machine learning in chemistry and materials science suffer from small dataset sizes, which demands special care for the model design in order to deliver reliable predictions. Hence, feature selection as the key determinant for dataset design is essential here. We propose a practical and efficient feature filter strategy to determine the best input feature candidates. We illustrate this strategy for the prediction of adsorption energies based on a public dataset and sublimation enthalpies using an in-house training dataset. The input of adsorption energies reduces the feature space from 12 dimensions to two and still delivers accurate results. For the sublimation enthalpies, three input configurations are filtered from 14 possible configurations with different dimensions for further productive predictions as being most relevant by using our feature filter strategy. The best extreme gradient boosting regression model possesses a good performance and is evaluated from statistical and theoretical perspectives, reaching a level of accuracy comparable to density functional theory computations and allowing for physical interpretations of the predictions. Overall, the results indicate that the feature filter strategy can help interdisciplinary scientists without rich professional AI knowledge and limited computational resources to establish a reliable small training dataset first, which may make the final machine learning model training easier and more accurate, avoiding time-consuming hyperparameter explorations and improper feature selection.

Enhanced motor noise in an autism subtype with poor motor skills

BackgroundMotor difficulties are common in many, but not all, autistic individuals. These difficulties can co-occur with other problems, such as delays in language, intellectual, and adaptive functioning. Biological mechanisms underpinning such difficulties are less well understood. Poor motor skills tend to be more common in individuals carrying highly penetrant rare genetic mutations. Such mechanisms may have downstream consequences of altering neurophysiological excitation-inhibition balance and lead to enhanced behavioral motor noise.MethodsThis study combined publicly available and in-house datasets of autistic (n = 156), typically-developing (TD, n = 149), and developmental coordination disorder (DCD, n = 23) children (age 3–16 years). Autism motor subtypes were identified based on patterns of motor abilities measured from the Movement Assessment Battery for Children 2nd edition. Stability-based relative clustering validation was used to identify autism motor subtypes and evaluate generalization accuracy in held-out data. Autism motor subtypes were tested for differences in motor noise, operationalized as the degree of dissimilarity between repeated motor kinematic trajectories recorded during a simple reach-to-drop task.ResultsRelatively ‘high’ (n = 87) versus ‘low’ (n = 69) autism motor subtypes could be detected and which generalize with 89% accuracy in held-out data. The relatively ‘low’ subtype was lower in general intellectual ability and older at age of independent walking, but did not differ in age at first words or autistic traits or symptomatology. Motor noise was considerably higher in the ‘low’ subtype compared to ‘high’ (Cohen’s d = 0.77) or TD children (Cohen’s d = 0.85), but similar between autism ‘high’ and TD children (Cohen’s d = 0.08). Enhanced motor noise in the ‘low’ subtype was also most pronounced during the feedforward phase of reaching actions.LimitationsThe sample size of this work is limited. Future work in larger samples along with independent replication is important. Motor noise was measured only on one specific motor task. Thus, a more comprehensive assessment of motor noise on many other motor tasks is needed.ConclusionsAutism can be split into at least two discrete motor subtypes that are characterized by differing levels of motor noise. This suggests that autism motor subtypes may be underpinned by different biological mechanisms.

Predicting sessile droplet evaporation kinetics via cascaded deep networks and tree-based machine learning approach

The intricate nature of sessile droplet evaporation phenomenon makes detailed experimental studies time consuming and requires sophisticated apparatus; while complete numerical simulations are computationally expensive and also time-intensive. In this article, for the first time, we explore the applicability of machine learning (ML) approaches to predict the evaporation kinetics of generalized sessile droplets under various conditions. An in-house dataset, obtained using an experimentally well validated numerical model, is used to develop the ML models: deep artificial neural network (ANN) and decision tree algorithms such as Random Forest (RF) and extreme gradient boosting (XGB). The structures of these models are modified by cascading the output features according to the physics involved. This distinctive approach results in better prediction of the evaporation kinetics than basic ML models. The models are trained by a large set of input parameters for the target variables: viz. droplet evaporation rate, velocity scale, and temperature drop in the solid and liquid domain. Finally, the performance of these ML models is assessed by comparing their predictions with that of physics-based, experimentally validated, numerical models. Results show that the inclusion of additional features obtained using feature engineering significantly improve the prediction performance of ML algorithms, and consistently accurate predictions of droplet evaporation kinetics are obtained. Among various algorithms considered here, the ANN outperforms in term of various error matrices for most of the cases, followed by XGB, and RF models. Also, the highest mean average error (MAE) yield by the ML models for evaporation rate, velocity scale and temperature drop in liquid remains within ∼12.5%. In the case of temperature drop for the solid, the MAE is considerably higher due to large variability of the same target variable. Overall, the work clearly shows that ML algorithms can be used to obtain physically consistent predictions for sessile droplet evaporation parameters.

Cross-Modal Consistency for Single-Modal MR Image Segmentation.

Multi-modal magnetic resonance (MR) image segmentation is an important task in disease diagnosis and treatment, but it is usually difficult to obtain multiple modalities for a single patient in clinical applications. To address these issues, a cross-modal consistency framework is proposed for a single-modal MR image segmentation. To enable single-modal MR image segmentation in the inference stage, a weighted cross-entropy loss and a pixel-level feature consistency loss are proposed to train the target network with the guidance of the teacher network and the auxiliary network. To fuse dual-modal MR images in the training stage, the cross-modal consistency is measured according to Dice similarity entropy loss and Dice similarity contrastive loss, so as to maximize the prediction similarity of the teacher network and the auxiliary network. To reduce the difference in image contrast between different MR images for the same organs, a contrast alignment network is proposed to align input images with different contrasts to reference images with a good contrast. Comprehensive experiments have been performed on a publicly available prostate dataset and an in-house pancreas dataset to verify the effectiveness of the proposed method. Compared to state-of-the-art methods, the proposed method can achieve better segmentation. The proposed image segmentation method can fuse dual-modal MR images in the training stage and only need one-modal MR images in the inference stage. The proposed method can be used in routine clinical occasions when only single-modal MR image with variable contrast is available for a patient.

Higher Order Polynomial Transformer for Fine-Grained Freezing of Gait Detection.

Freezing of Gait (FoG) is a common symptom of Parkinson's disease (PD), manifesting as a brief, episodic absence, or marked reduction in walking, despite a patient's intention to move. Clinical assessment of FoG events from manual observations by experts is both time-consuming and highly subjective. Therefore, machine learning-based FoG identification methods would be desirable. In this article, we address this task as a fine-grained human action recognition problem based on vision inputs. A novel deep learning architecture, namely, higher order polynomial transformer (HP-Transformer), is proposed to incorporate pose and appearance feature sequences to formulate fine-grained FoG patterns. In particular, a higher order self-attention mechanism is proposed based on higher order polynomials. To this end, linear, bilinear, and trilinear transformers are formulated in pursuit of discriminative fine-grained representations. These representations are treated as multiple streams and further fused by a cross-order fusion strategy for FoG detection. Comprehensive experiments on a large in-house dataset collected during clinical assessments demonstrate the effectiveness of the proposed method, and an area under the receiver operating characteristic (ROC) curve (AUC) of 0.92 is achieved for detecting FoG.