J Chem Inf Model. 2024 Jun 28. doi: 10.1021/acs.jcim.4c00432. Online ahead of print.

ABSTRACT

Computer-assisted synthesis planning has become increasingly important in drug discovery. While deep-learning models have shown remarkable progress in achieving high accuracies for single-step retrosynthetic predictions, their performances in retrosynthetic route planning need to be checked. This study compares the intricate single-step models with a straightforward template enumeration approach for retrosynthetic route planning on a real-world drug molecule data set. Despite the superior single-step accuracy of advanced models, the template enumeration method with a heuristic-based retrosynthesis knowledge score was found to surpass them in efficiency in searching the reaction space, achieving a higher or comparable solve rate within the same time frame. This counterintuitive result underscores the importance of efficiency and retrosynthesis knowledge in retrosynthesis route planning and suggests that future research should incorporate a simple template enumeration as a benchmark. It also suggests that this simple yet effective strategy should be considered alongside more complex models to better cater to the practical needs of computer-assisted synthesis planning in drug discovery.

PMID:38940765 | DOI:10.1021/acs.jcim.4c00432

Bioinformatics. 2024 Jun 28;40(Supplement_1):i169-i179. doi: 10.1093/bioinformatics/btae264.

ABSTRACT

MOTIVATION: Electronic health records (EHRs) represent a comprehensive resource of a patient's medical history. EHRs are essential for utilizing advanced technologies such as deep learning (DL), enabling healthcare providers to analyze extensive data, extract valuable insights, and make precise and data-driven clinical decisions. DL methods such as recurrent neural networks (RNN) have been utilized to analyze EHR to model disease progression and predict diagnosis. However, these methods do not address some inherent irregularities in EHR data such as irregular time intervals between clinical visits. Furthermore, most DL models are not interpretable. In this study, we propose two interpretable DL architectures based on RNN, namely time-aware RNN (TA-RNN) and TA-RNN-autoencoder (TA-RNN-AE) to predict patient's clinical outcome in EHR at the next visit and multiple visits ahead, respectively. To mitigate the impact of irregular time intervals, we propose incorporating time embedding of the elapsed times between visits. For interpretability, we propose employing a dual-level attention mechanism that operates between visits and features within each visit.

RESULTS: The results of the experiments conducted on Alzheimer's Disease Neuroimaging Initiative (ADNI) and National Alzheimer's Coordinating Center (NACC) datasets indicated the superior performance of proposed models for predicting Alzheimer's Disease (AD) compared to state-of-the-art and baseline approaches based on F2 and sensitivity. Additionally, TA-RNN showed superior performance on the Medical Information Mart for Intensive Care (MIMIC-III) dataset for mortality prediction. In our ablation study, we observed enhanced predictive performance by incorporating time embedding and attention mechanisms. Finally, investigating attention weights helped identify influential visits and features in predictions.

AVAILABILITY AND IMPLEMENTATION: https://github.com/bozdaglab/TA-RNN.

PMID:38940180 | DOI:10.1093/bioinformatics/btae264

Bioinformatics. 2024 Jun 28;40(Supplement_1):i539-i547. doi: 10.1093/bioinformatics/btae240.

ABSTRACT

MOTIVATION: In drug discovery, it is crucial to assess the drug-target binding affinity (DTA). Although molecular docking is widely used, computational efficiency limits its application in large-scale virtual screening. Deep learning-based methods learn virtual scoring functions from labeled datasets and can quickly predict affinity. However, there are three limitations. First, existing methods only consider the atom-bond graph or one-dimensional sequence representations of compounds, ignoring the information about functional groups (pharmacophores) with specific biological activities. Second, relying on limited labeled datasets fails to learn comprehensive embedding representations of compounds and proteins, resulting in poor generalization performance in complex scenarios. Third, existing feature fusion methods cannot adequately capture contextual interaction information.

RESULTS: Therefore, we propose a novel DTA prediction method named HeteroDTA. Specifically, a multi-view compound feature extraction module is constructed to model the atom-bond graph and pharmacophore graph. The residue concat graph and protein sequence are also utilized to model protein structure and function. Moreover, to enhance the generalization capability and reduce the dependence on task-specific labeled data, pre-trained models are utilized to initialize the atomic features of the compounds and the embedding representations of the protein sequence. A context-aware nonlinear feature fusion method is also proposed to learn interaction patterns between compounds and proteins. Experimental results on public benchmark datasets show that HeteroDTA significantly outperforms existing methods. In addition, HeteroDTA shows excellent generalization performance in cold-start experiments and superiority in the representation learning ability of drug-target pairs. Finally, the effectiveness of HeteroDTA is demonstrated in a real-world drug discovery study.

AVAILABILITY AND IMPLEMENTATION: The source code and data are available at https://github.com/daydayupzzl/HeteroDTA.

PMID:38940179 | DOI:10.1093/bioinformatics/btae240

Bioinformatics. 2024 Jun 28;40(Supplement_1):i357-i368. doi: 10.1093/bioinformatics/btae260.

ABSTRACT

MOTIVATION: The current paradigm of deep learning models for the joint representation of molecules and text primarily relies on 1D or 2D molecular formats, neglecting significant 3D structural information that offers valuable physical insight. This narrow focus inhibits the models' versatility and adaptability across a wide range of modalities. Conversely, the limited research focusing on explicit 3D representation tends to overlook textual data within the biomedical domain.

RESULTS: We present a unified pre-trained language model, MolLM, that concurrently captures 2D and 3D molecular information alongside biomedical text. MolLM consists of a text Transformer encoder and a molecular Transformer encoder, designed to encode both 2D and 3D molecular structures. To support MolLM's self-supervised pre-training, we constructed 160K molecule-text pairings. Employing contrastive learning as a supervisory signal for learning, MolLM demonstrates robust molecular representation capabilities across four downstream tasks, including cross-modal molecule and text matching, property prediction, captioning, and text-prompted molecular editing. Through ablation, we demonstrate that the inclusion of explicit 3D representations improves performance in these downstream tasks.

AVAILABILITY AND IMPLEMENTATION: Our code, data, pre-trained model weights, and examples of using our model are all available at https://github.com/gersteinlab/MolLM. In particular, we provide Jupyter Notebooks offering step-by-step guidance on how to use MolLM to extract embeddings for both molecules and text.

PMID:38940177 | DOI:10.1093/bioinformatics/btae260

Bioinformatics. 2024 Jun 28;40(Supplement_1):i91-i99. doi: 10.1093/bioinformatics/btae261.

ABSTRACT

MOTIVATION: High-throughput screens (HTS) provide a powerful tool to decipher the causal effects of chemical and genetic perturbations on cancer cell lines. Their ability to evaluate a wide spectrum of interventions, from single drugs to intricate drug combinations and CRISPR-interference, has established them as an invaluable resource for the development of novel therapeutic approaches. Nevertheless, the combinatorial complexity of potential interventions makes a comprehensive exploration intractable. Hence, prioritizing interventions for further experimental investigation becomes of utmost importance.

RESULTS: We propose CODEX (COunterfactual Deep learning for the in silico EXploration of cancer cell line perturbations) as a general framework for the causal modeling of HTS data, linking perturbations to their downstream consequences. CODEX relies on a stringent causal modeling strategy based on counterfactual reasoning. As such, CODEX predicts drug-specific cellular responses, comprising cell survival and molecular alterations, and facilitates the in silico exploration of drug combinations. This is achieved for both bulk and single-cell HTS. We further show that CODEX provides a rationale to explore complex genetic modifications from CRISPR-interference in silico in single cells.

AVAILABILITY AND IMPLEMENTATION: Our implementation of CODEX is publicly available at https://github.com/sschrod/CODEX. All data used in this article are publicly available.

PMID:38940173 | DOI:10.1093/bioinformatics/btae261

Bioinformatics. 2024 Jun 28;40(Supplement_1):i381-i389. doi: 10.1093/bioinformatics/btae262.

ABSTRACT

SUMMARY: Cis-acting mRNA elements play a key role in the regulation of mRNA stability and translation efficiency. Revealing the interactions of these elements and their impact plays a crucial role in understanding the regulation of the mRNA translation process, which supports the development of mRNA-based medicine or vaccines. Deep neural networks (DNN) can learn complex cis-regulatory codes from RNA sequences. However, extracting these cis-regulatory codes efficiently from DNN remains a significant challenge. Here, we propose a method based on our toolkit NeuronMotif and motif mutagenesis, which not only enables the discovery of diverse and high-quality motifs but also efficiently reveals motif interactions. By interpreting deep-learning models, we have discovered several crucial motifs that impact mRNA translation efficiency and stability, as well as some unknown motifs or motif syntax, offering novel insights for biologists. Furthermore, we note that it is challenging to enrich motif syntax in datasets composed of randomly generated sequences, and they may not contain sufficient biological signals.

AVAILABILITY AND IMPLEMENTATION: The source code and data used to produce the results and analyses presented in this manuscript are available from GitHub (https://github.com/WangLabTHU/combmotif).

PMID:38940172 | DOI:10.1093/bioinformatics/btae262

Bioinformatics. 2024 Jun 28;40(Supplement_1):i462-i470. doi: 10.1093/bioinformatics/btae248.

ABSTRACT

SUMMARY: Phenotype-based drug screening emerges as a powerful approach for identifying compounds that actively interact with cells. Transcriptional and proteomic profiling of cell lines and individual cells provide insights into the cellular state alterations that occur at the molecular level in response to external perturbations, such as drugs or genetic manipulations. In this paper, we propose cycleCDR, a novel deep learning framework to predict cellular response to external perturbations. We leverage the autoencoder to map the unperturbed cellular states to a latent space, in which we postulate the effects of drug perturbations on cellular states follow a linear additive model. Next, we introduce the cycle consistency constraints to ensure that unperturbed cellular state subjected to drug perturbation in the latent space would produces the perturbed cellular state through the decoder. Conversely, removal of perturbations from the perturbed cellular states can restore the unperturbed cellular state. The cycle consistency constraints and linear modeling in the latent space enable to learn transferable representations of external perturbations, so that our model can generalize well to unseen drugs during training stage. We validate our model on four different types of datasets, including bulk transcriptional responses, bulk proteomic responses, and single-cell transcriptional responses to drug/gene perturbations. The experimental results demonstrate that our model consistently outperforms existing state-of-the-art methods, indicating our method is highly versatile and applicable to a wide range of scenarios.

AVAILABILITY AND IMPLEMENTATION: The source code is available at: https://github.com/hliulab/cycleCDR.

PMID:38940153 | DOI:10.1093/bioinformatics/btae248

Bioinformatics. 2024 Jun 28;40(Supplement_1):i490-i500. doi: 10.1093/bioinformatics/btae223.

ABSTRACT

SUMMARY: Single-cell Hi-C (scHi-C) protocol helps identify cell-type-specific chromatin interactions and sheds light on cell differentiation and disease progression. Despite providing crucial insights, scHi-C data is often underutilized due to the high cost and the complexity of the experimental protocol. We present a deep learning framework, scGrapHiC, that predicts pseudo-bulk scHi-C contact maps using pseudo-bulk scRNA-seq data. Specifically, scGrapHiC performs graph deconvolution to extract genome-wide single-cell interactions from a bulk Hi-C contact map using scRNA-seq as a guiding signal. Our evaluations show that scGrapHiC, trained on seven cell-type co-assay datasets, outperforms typical sequence encoder approaches. For example, scGrapHiC achieves a substantial improvement of 23.2% in recovering cell-type-specific Topologically Associating Domains over the baselines. It also generalizes to unseen embryo and brain tissue samples. scGrapHiC is a novel method to generate cell-type-specific scHi-C contact maps using widely available genomic signals that enables the study of cell-type-specific chromatin interactions.

AVAILABILITY AND IMPLEMENTATION: The GitHub link: https://github.com/rsinghlab/scGrapHiC contains the source code of scGrapHiC and associated scripts to preprocess publicly available datasets to produce the results and visualizations we have discuss in this manuscript.

PMID:38940151 | DOI:10.1093/bioinformatics/btae223

Bioinformatics. 2024 Jun 28;40(Supplement_1):i110-i118. doi: 10.1093/bioinformatics/btae239.

ABSTRACT

Artificial intelligence (AI) is increasingly used in genomics research and practice, and generative AI has garnered significant recent attention. In clinical applications of generative AI, aspects of the underlying datasets can impact results, and confounders should be studied and mitigated. One example involves the facial expressions of people with genetic conditions. Stereotypically, Williams (WS) and Angelman (AS) syndromes are associated with a "happy" demeanor, including a smiling expression. Clinical geneticists may be more likely to identify these conditions in images of smiling individuals. To study the impact of facial expression, we analyzed publicly available facial images of approximately 3500 individuals with genetic conditions. Using a deep learning (DL) image classifier, we found that WS and AS images with non-smiling expressions had significantly lower prediction probabilities for the correct syndrome labels than those with smiling expressions. This was not seen for 22q11.2 deletion and Noonan syndromes, which are not associated with a smiling expression. To further explore the effect of facial expressions, we computationally altered the facial expressions for these images. We trained HyperStyle, a GAN-inversion technique compatible with StyleGAN2, to determine the vector representations of our images. Then, following the concept of InterfaceGAN, we edited these vectors to recreate the original images in a phenotypically accurate way but with a different facial expression. Through online surveys and an eye-tracking experiment, we examined how altered facial expressions affect the performance of human experts. We overall found that facial expression is associated with diagnostic accuracy variably in different genetic conditions.

PMID:38940144 | DOI:10.1093/bioinformatics/btae239

Bioinformatics. 2024 Jun 28;40(Supplement_1):i257-i265. doi: 10.1093/bioinformatics/btae220.

ABSTRACT

MOTIVATION: Tandem mass spectrometry (MS/MS) is a crucial technology for large-scale proteomic analysis. The protein database search or the spectral library search are commonly used for peptide identification from MS/MS spectra, which, however, may face challenges due to experimental variations between replicated spectra and similar fragmentation patterns among distinct peptides. To address this challenge, we present SpecEncoder, a deep metric learning approach to address these challenges by transforming MS/MS spectra into robust and sensitive embedding vectors in a latent space. The SpecEncoder model can also embed predicted MS/MS spectra of peptides, enabling a hybrid search approach that combines spectral library and protein database searches for peptide identification.

RESULTS: We evaluated SpecEncoder on three large human proteomics datasets, and the results showed a consistent improvement in peptide identification. For spectral library search, SpecEncoder identifies 1%-2% more unique peptides (and PSMs) than SpectraST. For protein database search, it identifies 6%-15% more unique peptides than MSGF+ enhanced by Percolator, Furthermore, SpecEncoder identified 6%-12% additional unique peptides when utilizing a combined library of experimental and predicted spectra. SpecEncoder can also identify more peptides when compared to deep-learning enhanced methods (MSFragger boosted by MSBooster). These results demonstrate SpecEncoder's potential to enhance peptide identification for proteomic data analyses.

AVAILABILITY AND IMPLEMENTATION: The source code and scripts for SpecEncoder and peptide identification are available on GitHub at https://github.com/lkytal/SpecEncoder. Contact: hatang@iu.edu.

PMID:38940141 | DOI:10.1093/bioinformatics/btae220

Bioinformatics. 2024 Jun 28;40(Supplement_1):i180-i188. doi: 10.1093/bioinformatics/btae214.

ABSTRACT

MOTIVATION: Understanding the genomic heterogeneity of tumors is an important task in computational oncology, especially in the context of finding personalized treatments based on the genetic profile of each patient's tumor. Tumor clustering that takes into account the temporal order of genetic events, as represented by tumor mutation trees, is a powerful approach for grouping together patients with genetically and evolutionarily similar tumors and can provide insights into discovering tumor subtypes, for more accurate clinical diagnosis and prognosis.

RESULTS: Here, we propose oncotree2vec, a method for clustering tumor mutation trees by learning vector representations of mutation trees that capture the different relationships between subclones in an unsupervised manner. Learning low-dimensional tree embeddings facilitates the visualization of relations between trees in large cohorts and can be used for downstream analyses, such as deep learning approaches for single-cell multi-omics data integration. We assessed the performance and the usefulness of our method in three simulation studies and on two real datasets: a cohort of 43 trees from six cancer types with different branching patterns corresponding to different modes of spatial tumor evolution and a cohort of 123 AML mutation trees.

AVAILABILITY AND IMPLEMENTATION: https://github.com/cbg-ethz/oncotree2vec.

PMID:38940124 | DOI:10.1093/bioinformatics/btae214

J Imaging Inform Med. 2024 Jun 27. doi: 10.1007/s10278-024-01175-x. Online ahead of print.

ABSTRACT

Spine disorders can cause severe functional limitations, including back pain, decreased pulmonary function, and increased mortality risk. Plain radiography is the first-line imaging modality to diagnose suspected spine disorders. Nevertheless, radiographical appearance is not always sufficient due to highly variable patient and imaging parameters, which can lead to misdiagnosis or delayed diagnosis. Employing an accurate automated detection model can alleviate the workload of clinical experts, thereby reducing human errors, facilitating earlier detection, and improving diagnostic accuracy. To this end, deep learning-based computer-aided diagnosis (CAD) tools have significantly outperformed the accuracy of traditional CAD software. Motivated by these observations, we proposed a deep learning-based approach for end-to-end detection and localization of spine disorders from plain radiographs. In doing so, we took the first steps in employing state-of-the-art transformer networks to differentiate images of multiple spine disorders from healthy counterparts and localize the identified disorders, focusing on vertebral compression fractures (VCF) and spondylolisthesis due to their high prevalence and potential severity. The VCF dataset comprised 337 images, with VCFs collected from 138 subjects and 624 normal images collected from 337 subjects. The spondylolisthesis dataset comprised 413 images, with spondylolisthesis collected from 336 subjects and 782 normal images collected from 413 subjects. Transformer-based models exhibited 0.97 Area Under the Receiver Operating Characteristic Curve (AUC) in VCF detection and 0.95 AUC in spondylolisthesis detection. Further, transformers demonstrated significant performance improvements against existing end-to-end approaches by 4-14% AUC (p-values < 10-13) for VCF detection and by 14-20% AUC (p-values < 10-9) for spondylolisthesis detection.

PMID:38937344 | DOI:10.1007/s10278-024-01175-x

J Imaging Inform Med. 2024 Jun 27. doi: 10.1007/s10278-024-01173-z. Online ahead of print.

ABSTRACT

As the adoption of artificial intelligence (AI) systems in radiology grows, the increase in demand for greater bandwidth and computational resources can lead to greater infrastructural costs for healthcare providers and AI vendors. To that end, we developed ISLE, an intelligent streaming framework to address inefficiencies in current imaging infrastructures. Our framework draws inspiration from video-on-demand platforms to intelligently stream medical images to AI vendors at an optimal resolution for inference from a single high-resolution copy using progressive encoding. We hypothesize that ISLE can dramatically reduce the bandwidth and computational requirements for AI inference, while increasing throughput (i.e., the number of scans processed by the AI system per second). We evaluate our framework by streaming chest X-rays for classification and abdomen CT scans for liver and spleen segmentation and comparing them with the original versions of each dataset. For classification, our results show that ISLE reduced data transmission and decoding time by at least 92% and 88%, respectively, while increasing throughput by more than 3.72 × . For both segmentation tasks, ISLE reduced data transmission and decoding time by at least 82% and 88%, respectively, while increasing throughput by more than 2.9 × . In all three tasks, the ISLE streamed data had no impact on the AI system's diagnostic performance (all P > 0.05). Therefore, our results indicate that our framework can address inefficiencies in current imaging infrastructures by improving data and computational efficiency of AI deployments in the clinical environment without impacting clinical decision-making using AI systems.

PMID:38937343 | DOI:10.1007/s10278-024-01173-z

J Imaging Inform Med. 2024 Jun 27. doi: 10.1007/s10278-024-01114-w. Online ahead of print.

ABSTRACT

Early and accurate detection of cervical lymph nodes is essential for the optimal management and staging of patients with head and neck malignancies. Pilot studies have demonstrated the potential for radiomic and artificial intelligence (AI) approaches in increasing diagnostic accuracy for the detection and classification of lymph nodes, but implementation of many of these approaches in real-world clinical settings would necessitate an automated lymph node segmentation pipeline as a first step. In this study, we aim to develop a non-invasive deep learning (DL) algorithm for detecting and automatically segmenting cervical lymph nodes in 25,119 CT slices from 221 normal neck contrast-enhanced CT scans from patients without head and neck cancer. We focused on the most challenging task of segmentation of small lymph nodes, evaluated multiple architectures, and employed U-Net and our adapted spatial context network to detect and segment small lymph nodes measuring 5-10 mm. The developed algorithm achieved a Dice score of 0.8084, indicating its effectiveness in detecting and segmenting cervical lymph nodes despite their small size. A segmentation framework successful in this task could represent an essential initial block for future algorithms aiming to evaluate small objects such as lymph nodes in different body parts, including small lymph nodes looking normal to the naked human eye but harboring early nodal metastases.

PMID:38937342 | DOI:10.1007/s10278-024-01114-w

Skeletal Radiol. 2024 Jun 27. doi: 10.1007/s00256-024-04733-0. Online ahead of print.

ABSTRACT

OBJECTIVE: To develop a whole-body low-dose CT (WBLDCT) deep learning model and determine its accuracy in predicting the presence of cytogenetic abnormalities in multiple myeloma (MM).

MATERIALS AND METHODS: WBLDCTs of MM patients performed within a year of diagnosis were included. Cytogenetic assessments of clonal plasma cells via fluorescent in situ hybridization (FISH) were used to risk-stratify patients as high-risk (HR) or standard-risk (SR). Presence of any of del(17p), t(14;16), t(4;14), and t(14;20) on FISH was defined as HR. The dataset was evenly divided into five groups (folds) at the individual patient level for model training. Mean and standard deviation (SD) of the area under the receiver operating curve (AUROC) across the folds were recorded.

RESULTS: One hundred fifty-one patients with MM were included in the study. The model performed best for t(4;14), mean (SD) AUROC of 0.874 (0.073). The lowest AUROC was observed for trisomies: AUROC of 0.717 (0.058). Two- and 5-year survival rates for HR cytogenetics were 87% and 71%, respectively, compared to 91% and 79% for SR cytogenetics. Survival predictions by the WBLDCT deep learning model revealed 2- and 5-year survival rates for patients with HR cytogenetics as 87% and 71%, respectively, compared to 92% and 81% for SR cytogenetics.

CONCLUSION: A deep learning model trained on WBLDCT scans predicted the presence of cytogenetic abnormalities used for risk stratification in MM. Assessment of the model's performance revealed good to excellent classification of the various cytogenetic abnormalities.

PMID:38937291 | DOI:10.1007/s00256-024-04733-0

Physiol Meas. 2024 Jun 27. doi: 10.1088/1361-6579/ad5cc0. Online ahead of print.

ABSTRACT

This study aims to address the challenges of imbalanced heartbeat classification using electrocardiogram (ECG). In this proposed novel deep-learning method, the focus is on accurately identifying minority classes in conditions characterized by significant imbalances in ECG data.&#xD; &#xD;Approach: We propose a Feature Fusion Neural Network enhanced by a Dynamic Minority-Biased Batch Weighting Loss Function. This network comprises three specialized branches: the Complete ECG Data Branch for a comprehensive view of ECG signals, the Local QRS Wave Branch for detailed features of the QRS complex, and the R Wave Information Branch to analyze R wave characteristics. This structure is designed to extract diverse aspects of ECG data. The dynamic loss function prioritizes minority classes while maintaining the recognition of majority classes, adjusting the network's learning focus without altering the original data distribution. Together, this fusion structure and adaptive loss function significantly improve the network's ability to distinguish between various heartbeat classes, enhancing the accuracy of minority class identification.&#xD;&#xD;Main Results: The proposed method demonstrated balanced performance within the MIT-BIH dataset, especially for minority classes. Under the intra-patient paradigm, the accuracy, sensitivity, specificity, and positive predictive value (PPV) for Supraventricular ectopic beat were 99.63%, 93.62%, 99.81%, and 92.98%, respectively, and for Fusion beat were 99.76%, 85.56%, 99.87%, and 84.16%, respectively. Under the inter-patient paradigm, these metrics were 96.56%, 89.16%, 96.84%, and 51.99% for Supraventricular ectopic beat, and 96.10%, 77.06%, 96.25%, and 13.92% for Fusion beat, respectively.&#xD;&#xD;Significance: This method effectively addresses the class imbalance in ECG datasets. By leveraging diverse ECG signal information and a novel loss function, this approach offers a promising tool for aiding in the diagnosis and treatment of cardiac conditions.&#xD.

PMID:38936397 | DOI:10.1088/1361-6579/ad5cc0

Med Image Anal. 2024 Jun 15;97:103239. doi: 10.1016/j.media.2024.103239. Online ahead of print.

ABSTRACT

In real-world clinical settings, traditional deep learning-based classification methods struggle with diagnosing newly introduced disease types because they require samples from all disease classes for offline training. Class incremental learning offers a promising solution by adapting a deep network trained on specific disease classes to handle new diseases. However, catastrophic forgetting occurs, decreasing the performance of earlier classes when adapting the model to new data. Prior proposed methodologies to overcome this require perpetual storage of previous samples, posing potential practical concerns regarding privacy and storage regulations in healthcare. To this end, we propose a novel data-free class incremental learning framework that utilizes data synthesis on learned classes instead of data storage from previous classes. Our key contributions include acquiring synthetic data known as Continual Class-Specific Impression (CCSI) for previously inaccessible trained classes and presenting a methodology to effectively utilize this data for updating networks when introducing new classes. We obtain CCSI by employing data inversion over gradients of the trained classification model on previous classes starting from the mean image of each class inspired by common landmarks shared among medical images and utilizing continual normalization layers statistics as a regularizer in this pixel-wise optimization process. Subsequently, we update the network by combining the synthesized data with new class data and incorporate several losses, including an intra-domain contrastive loss to generalize the deep network trained on the synthesized data to real data, a margin loss to increase separation among previous classes and new ones, and a cosine-normalized cross-entropy loss to alleviate the adverse effects of imbalanced distributions in training data. Extensive experiments show that the proposed framework achieves state-of-the-art performance on four of the public MedMNIST datasets and in-house echocardiography cine series, with an improvement in classification accuracy of up to 51% compared to baseline data-free methods. Our code is available at https://github.com/ubc-tea/Continual-Impression-CCSI.

PMID:38936223 | DOI:10.1016/j.media.2024.103239

Med Image Anal. 2024 Jun 15;97:103222. doi: 10.1016/j.media.2024.103222. Online ahead of print.

ABSTRACT

Since the rise of deep learning, new medical segmentation methods have rapidly been proposed with extremely promising results, often reporting marginal improvements on the previous state-of-the-art (SOTA) method. However, on visual inspection errors are often revealed, such as topological mistakes (e.g. holes or folds), that are not detected using traditional evaluation metrics. Incorrect topology can often lead to errors in clinically required downstream image processing tasks. Therefore, there is a need for new methods to focus on ensuring segmentations are topologically correct. In this work, we present TEDS-Net: a segmentation network that preserves anatomical topology whilst maintaining segmentation performance that is competitive with SOTA baselines. Further, we show how current SOTA segmentation methods can introduce problematic topological errors. TEDS-Net achieves anatomically plausible segmentation by using learnt topology-preserving fields to deform a prior. Traditionally, topology-preserving fields are described in the continuous domain and begin to break down when working in the discrete domain. Here, we introduce additional modifications that more strictly enforce topology preservation. We illustrate our method on an open-source medical heart dataset, performing both single and multi-structure segmentation, and show that the generated fields contain no folding voxels, which corresponds to full topology preservation on individual structures whilst vastly outperforming the other baselines on overall scene topology. The code is available at: https://github.com/mwyburd/TEDS-Net.

PMID:38936222 | DOI:10.1016/j.media.2024.103222

Biosens Bioelectron. 2024 Jun 25;262:116525. doi: 10.1016/j.bios.2024.116525. Online ahead of print.

ABSTRACT

Research has shown that plants have the ability to detect environmental changes and generate electrical signals in response. These electrical signals can regulate the physiological state of plants and produce corresponding feedback. This suggests that plants have the potential to be used as biosensors for monitoring environmental information. However, there are current challenges in linking environmental information with plant electrical signals, especially in collecting and classifying the corresponding electrical signals under soil moisture gradients. This study documented the electrical signals of clivia under different soil moisture gradients and created a dataset for classifying electrical signals. Subsequently, we proposed a lightweight convolutional neural network (CNN) model (PlantNet) for classifying the electrical signal dataset. Compared to traditional CNN models, our model achieved optimal classification performance with the lowest computational resource consumption. The model achieved an accuracy of 99.26%, precision of 99.31%, recall of 92.26%, F1-score of 99.21%, with 0.17M parameters, a size of 7.17MB, and 14.66M FLOPs. Therefore, this research provides scientific evidence for the future development of plants as biosensors for detecting soil moisture, and offers insight into developing plants as biosensors for detecting signals such as ozone, PM2.5, Volatile Organic Compounds(VOCs), and more. These studies are expected to drive the development of environmental monitoring technology and provide new pathways for better understanding the interaction between plants and the environment.

PMID:38936168 | DOI:10.1016/j.bios.2024.116525

Parasit Vectors. 2024 Jun 27;17(1):275. doi: 10.1186/s13071-024-06351-0.

ABSTRACT

BACKGROUND: Digital imaging combined with deep-learning-based computational image analysis is a growing area in medical diagnostics, including parasitology, where a number of automated analytical devices have been developed and are available for use in clinical practice.

METHODS: The performance of Parasight All-in-One (AIO), a second-generation device, was evaluated by comparing it to a well-accepted research method (mini-FLOTAC) and to another commercially available test (Imagyst). Fifty-nine canine and feline infected fecal specimens were quantitatively analyzed by all three methods. Since some samples were positive for more than one parasite, the dataset consisted of 48 specimens positive for Ancylostoma spp., 13 for Toxocara spp. and 23 for Trichuris spp.

RESULTS: The magnitude of Parasight AIO counts correlated well with those of mini-FLOTAC but not with those of Imagyst. Parasight AIO counted approximately 3.5-fold more ova of Ancylostoma spp. and Trichuris spp. and 4.6-fold more ova of Toxocara spp. than the mini-FLOTAC, and counted 27.9-, 17.1- and 10.2-fold more of these same ova than Imagyst, respectively. These differences translated into differences between the test sensitivities at low egg count levels (< 50 eggs/g), with Parasight AIO > mini-FLOTAC > Imagyst. At higher egg counts Parasight AIO and mini-FLOTAC performed with comparable precision (which was significantly higher that than Imagyst), whereas at lower counts (> 30 eggs/g) Parasight was more precise than both mini-FLOTAC and Imagyst, while the latter two methods did not significantly differ from each other.

CONCLUSIONS: In general, Parasight AIO analyses were both more precise and sensitive than mini-FLOTAC and Imagyst and quantitatively correlated well with mini-FLOTAC. While Parasight AIO produced lower raw counts in eggs-per-gram than mini-FLOTAC, these could be corrected using the data generated from these correlations.

PMID:38937854 | DOI:10.1186/s13071-024-06351-0