ß-Sitosterol alleviates the malignant phenotype of hepatocellular carcinoma cells via inhibiting GSK3B expression.

To explore the effects of ß-Sitosterol upon hepatocellular carcinoma cell proliferation, apoptosis, migration, invasion, and epithelial-mesenchymal transition (EMT), and to investigate the underlying mechanism using network pharmacology. Human hepatocellular carcinoma cell lines (Huh-7 and HCCLM3) were expose to gradient concentrations of ß-Sitosterol (5 µg/mL, 10 µg/mL, and 20 µg/mL). Cell viability and proliferation were assessed using MTT, CCK-8, colony formation, and EdU assays.Flow cytometry was employed to evaluate cell cycle and apoptosis. Scratch and Transwell assays were performed, respectively, to detect cell migration and invasion. The levels of apoptosis-associated proteins (BAX, BCL2, and cleaved caspase3) as well as EMT-associated proteins (E-cadherin, N-cadherin, Snail, and Vimentin) were detected in Huh-7 and HCCLM3 cell lines using Western blot analysis. The drug target gene for ß-Sitosterol was screened via PubChem and subsequently evaluated for expression in the GSE112790 dataset. In addition, the expression level of glycogen synthase kinase 3 beta (GSK3B) within the Cancer Genome Atlas-Liver Hepatocellular Carcinoma (TCGA-LIHC) database was analyzed, along with its correlation to the survival outcomes of patients with hepatocellular carcinoma. The diagnostic efficiency of GSK3B was assessed by analyzing the ROC curve. Subsequently, Huh-7 and HCCLM3 cell lines were transfected with the overexpression vector of GSK3B and then treated with ß-Sitosterol to further validate the association between GSK3B and ß-Sitosterol. GSK3B demonstrated a significantly elevated expression in patients with hepatocellular carcinoma, which could predict hepatocellular carcinoma patients' impaired prognosis based on GEO dataset and TCGA database. GSK3B inhibitor (CHIR-98014) notably inhibited cell proliferation and invasion, promoted cell apoptosis and cell cycle arrest at G0/G1 phase in hepatocellular carcinoma cells. ß-Sitosterol treatment further promoted the efffects of GSK3B inhibitor on hepatocellular carcinoma cells. GSK3B overexpression has been found to enhance the proliferative and invasive capabilities of hepatocellular carcinoma cells. Furthermore it has been observed that GSK3B overexpression, it has been obsear can partially reverse the inhibitory effect of ß-Sitosterol upon hepatocellular. ß-Sitosterol suppressed hepatocellular carcinoma cell proliferation and invasion, and enhanced apoptosis via inhibiting GSK3B expression.

Achievement of efficient thermally activated delayed fluorescence materials based on 1,8-naphthalimide derivatives exhibiting piezochromic and thermochromic luminescence.

In this study, we developed a D-A type imide derivative based on 1,8-naphthalimide, NI-mPCz, which exhibited outstanding thermally activated delayed fluorescence (TADF) properties. Additionally, it demonstrates characteristics of piezochromic and thermochromic luminescence. The thermochromic luminescence observed is attributed to crystalline transformations occurring during the heating process, as evidenced by differential scanning calorimetry (DSC) and microscopic examinations. Moreover, the good compatibility of NI-mPCz with HeLa cells and its excellent imaging performance indicate its potential for application in the field of biological imaging. These results provide valuable insights for the design and development of new organic electronic and bioimaging materials with high-efficiency TADF characteristics.

Data leakage in deep learning studies of translational EEG.

A growing number of studies apply deep neural networks (DNNs) to recordings of human electroencephalography (EEG) to identify a range of disorders. In many studies, EEG recordings are split into segments, and each segment is randomly assigned to the training or test set. As a consequence, data from individual subjects appears in both the training and the test set. Could high test-set accuracy reflect data leakage from subject-specific patterns in the data, rather than patterns that identify a disease? We address this question by testing the performance of DNN classifiers using segment-based holdout (in which segments from one subject can appear in both the training and test set), and comparing this to their performance using subject-based holdout (where all segments from one subject appear exclusively in either the training set or the test set). In two datasets (one classifying Alzheimer's disease, and the other classifying epileptic seizures), we find that performance on previously-unseen subjects is strongly overestimated when models are trained using segment-based holdout. Finally, we survey the literature and find that the majority of translational DNN-EEG studies use segment-based holdout. Most published DNN-EEG studies may dramatically overestimate their classification performance on new subjects.

Usformer: A small network for left atrium segmentation of 3D LGE MRI.

Left atrial (LA) fibrosis plays a vital role as a mediator in the progression of atrial fibrillation. 3D late gadolinium-enhancement (LGE) MRI has been proven effective in identifying LA fibrosis. Image analysis of 3D LA LGE involves manual segmentation of the LA wall, which is both lengthy and challenging. Automated segmentation poses challenges owing to the diverse intensities in data from various vendors, the limited contrast between LA and surrounding tissues, and the intricate anatomical structures of the LA. Current approaches relying on 3D networks are computationally intensive since 3D LGE MRIs and the networks are large. Regarding this issue, most researchers came up with two-stage methods: initially identifying the LA center using a scaled-down version of the MRIs and subsequently cropping the full-resolution MRIs around the LA center for final segmentation. We propose a lightweight transformer-based 3D architecture, Usformer, designed to precisely segment LA volume in a single stage, eliminating error propagation associated with suboptimal two-stage training. The transposed attention facilitates capturing the global context in large 3D volumes without significant computation requirements. Usformer outperforms the state-of-the-art supervised learning methods in terms of accuracy and speed. First, with the smallest Hausdorff Distance (HD) and Average Symmetric Surface Distance (ASSD), it achieved a dice score of 93.1% and 92.0% in the 2018 Atrial Segmentation Challenge and our local institutional dataset, respectively. Second, the number of parameters and computation complexity are largely reduced by 2.8x and 3.8x, respectively. Moreover, Usformer does not require a large dataset. When only 16 labeled MRI scans are used for training, Usformer achieves a 92.1% dice score in the challenge dataset. The proposed Usformer delineates the boundaries of the LA wall relatively accurately, which may assist in the clinical translation of LA LGE for planning catheter ablation of atrial fibrillation.

Efficient orange and red thermally activated delayed fluorescence materials based on 1,8-naphthalimide derivatives.

Thermally activated delayed fluorescence (TADF) molecules have emerged as a promising class of third-generation organic light-emitting diode (OLED) emitters that can achieve 100% internal quantum efficiency without the use of noble metals. However, the design of high-efficiency red TADF materials has been challenging due to limitations imposed by the energy-gap law. To overcome this challenge, two new TADF emitters, namely, 6-(4-(diphenylamino)phenyl)-2-phenyl-1H-benzo[de]isoquinoline-1,3(2H)-dione (NI-TPA) and 6-(10H-phenothiazin-10-yl)-2-phenyl-1H-benzo[de]-isoquinoline-1,3(2H)-dione (NI-Pz), have been synthesized and characterized. These compounds exhibit strong TADF characteristics with a small energy gap (ΔEST) between the lowest excited singlet and triplet states, short delayed fluorescence lifetimes, high thermal stability, and high photoluminescence quantum yields. The OLED devices fabricated using NI-TPA and NI-Pz as emitters show orange and red electroluminescence with emission peaks at 593 nm and 665 nm, respectively, and maximum external quantum efficiencies (EQEs) of 11.3% and 7.6%, respectively. Furthermore, applying NI-TPA to cell imaging yielded excellent imaging results, indicating the potential of red TADF materials in the field of biological imaging.

Complexes of glycated casein and carboxymethyl cellulose enhance stability and control release of anthocyanins.

To improve the stability and sustained-release property of anthocyanins (ACNs), casein (CA) - dextran (DEX) glycated conjugates (UGCA) and carboxymethyl cellulose (CMC) were used to prepare ACNs-loaded binary and ternary complexes. The ACNs-loaded binary complexes (ACNs-UGCA) and ternary complexes (ACNs-UGCA-CMC) achieved by 8 min' ultrasonic treatment with 40 % amplitude. The binary and ternary complexes showed spherical structure and good dispersibility, with the average size of 121.2 nm and 132.4 nm respectively. The anthocyanins encapsulation efficiency of ACNs-UGCA-CMC increased almost 20 % than ACNs-UGCA. ACNs-UGCA-CMC had better colloidal stabilities than ACNs-UGCA, such as thermal stability and dilution stability. Simultaneously, both of the binary and ternary complexes significantly prevented anthocyanins from being degraded by heat treatment, ascorbic acid, sucrose and simulated gastrointestinal environment. The protective effect of ACNs-UGCA-CMC was more significant. Furthermore, ACNs-UGCA-CMC showed slower anthocyanins release in simulated releasing environment in vitro and a long retention time in vivo. Our current study provides a potential delivery for improving the stability and controlling release of anthocyanins.

Multilevel holographic encryption based on the Tiger Amulet concept.

Optical holographic encryption (OHE) has been extensively researched in the field of information security due to its parallel and multi-dimensional characteristics. However, although some progress in OHE has been made in recent years, inherent security flaws resulting from the robust nature of holograms persist. In this study, we propose a multilevel holographic encryption method based on the Tiger Amulet (TA) concept. Compared with the normal OHE, our method employs two ciphertexts. It strategically utilizes the low-level plaintext as intentional deceptive content to confound the potential eavesdroppers. Furthermore, we ingeniously exploit the hologram's robustness in reverse, thereby establishing an additional protection mechanism to enhance the security of the middle-level plaintext. Leveraging the TA concept, the high-level plaintext can only be decrypted when two matched ciphertexts are combined and collimated. The TA based decryption mechanism enhances the security and sensitivity deciphering high-level plaintext. Benefiting from the security mechanisms above, our proposed method demonstrates promising applicability across diverse scenarios and holds the potential to redefine the landscape of multilevel OHE design.

Dual-function ß-cyclodextrin/starch-based intelligent film with reversible responsiveness and sustained bacteriostat-releasing for food preservation and monitoring.

The full combination of high sensitivity indication and long-lasting bacteriostatic function is an innovative need to meet the practicality of intelligent film packaging systems for food products. Hence, Blueberry anthocyanins (BA) copigmentated by ferulic acid (FA) was used as an indicator, and cinnamon essential oil (CO) encapsulated by ß-cyclodextrin (ß-CD) as a bacteriostat, potato starch (PS) as a film-forming substrate to prepared a dual-function starch-based intelligent active packaging film with pH indicator and antibacterial function. FA had the best copigmentation effect with a threefold increase in a value compared to other phenolic acids. The ΔE value increased from 3.24 to 5.13 at pH 2-8, and the change was still prominent in acid-base alternating test, indicating a high response sensitivity. Notably, the yellow gamut of indicating terminus increased its visibility to the naked eye. The release behavior of CO from film was in line with Fick's diffusion. Meanwhile, the release of CO delayed to about 90 h through ß-cyclodextrin encapsulation, showing a high growth-inhibition rate in E. coli and S. aureus of almost 100 %. In this study, a dual-function film with indication and bacteriostasis was prepared and enhanced with both, expanding its wide application in intelligent packaging of fresh food.

DBPFNet: a dual-band polarization image fusion network based on the attention mechanism and atrous spatial pyramid pooling.

In the current image fusion techniques, typically dual-band images are fused to obtain a fused image with salient target information, or intensity and polarization images are fused to achieve an image with enhanced visual perception. However, the current lack of dual-band polarization image datasets and effective fusion methods pose significant challenges for extracting more information in a single image. To address these problems, we construct a dataset containing intensity and polarization images in the visible and near-infrared bands. Furthermore, we propose an end-to-end image fusion network using attention mechanisms and atrous spatial pyramid pooling to extract key information and multi-scale global contextual information. Moreover, we design efficient loss functions to train the network. The experiments verify that the proposed method achieves better performance than the state-of-the-art in both subjective and objective evaluations.

Multi-key optical encryption based on two-channel incoherent scattering imaging.

Optical encryption has been extensively researched in the field of information security due to its characteristics of being parallel and multi-dimensionsal. However, most of the proposed multiple-image encryption systems suffer from a cross-talk problem. Here, we propose a multi-key optical encryption method based on a two-channel incoherent scattering imaging. In the encryption process, plaintexts are coded by the random phase mask (RPM) in each channel and then coupled by an incoherent superposition to form the output ciphertexts. In the decryption process, the plaintexts, keys, and ciphertexts, are treated as a system of two linear equations with two unknowns. By utilizing the principles of linear equations, the issue of cross-talk can be mathematically resolved. The proposed method enhances the security of the cryptosystem through the quantity and order of the keys. Specifically, the key space is significantly expanded by removing the requirement of uncorrected keys. This approach provides a superior method that can be easily implemented in various application scenarios.

DeepCOVID-Fuse: A Multi-Modality Deep Learning Model Fusing Chest X-rays and Clinical Variables to Predict COVID-19 Risk Levels.

The COVID-19 pandemic has posed unprecedented challenges to global healthcare systems, highlighting the need for accurate and timely risk prediction models that can prioritize patient care and allocate resources effectively. This study presents DeepCOVID-Fuse, a deep learning fusion model that predicts risk levels in patients with confirmed COVID-19 by combining chest radiographs (CXRs) and clinical variables. The study collected initial CXRs, clinical variables, and outcomes (i.e., mortality, intubation, hospital length of stay, Intensive care units (ICU) admission) from February to April 2020, with risk levels determined by the outcomes. The fusion model was trained on 1657 patients (Age: 58.30 ± 17.74; Female: 807) and validated on 428 patients (56.41 ± 17.03; 190) from the local healthcare system and tested on 439 patients (56.51 ± 17.78; 205) from a different holdout hospital. The performance of well-trained fusion models on full or partial modalities was compared using DeLong and McNemar tests. Results show that DeepCOVID-Fuse significantly (p < 0.05) outperformed models trained only on CXRs or clinical variables, with an accuracy of 0.658 and an area under the receiver operating characteristic curve (AUC) of 0.842. The fusion model achieves good outcome predictions even when only one of the modalities is used in testing, demonstrating its ability to learn better feature representations across different modalities during training.

Local super-resolution imaging of foveated areas in super-oscillating optical fields.

Herein, we propose a super-oscillation optical field foveated local super-resolution imaging method. Firstly, the post-diffraction integral equation of the foveated modulation device is constructed, the objective function and constraints are established, and the structural parameters of the amplitude modulation device are optimally solved by using genetic algorithm. Secondly, the solved data have been input into the software for point diffusion function analysis. We have studied the super-resolution performance of different ring band amplitude types, and find the 8-ring 0-1 amplitude type has the best super-resolution performance. Finally, the principle experimental device is built according to the simulation parameters, and the super-oscillatory device parameters is loaded onto the amplitude type spatial light modulator for the principle experiments, in which the super-oscillation foveated local super-resolution imaging system is able to perform high image contrast imaging in the whole field of view and super-resolution imaging in the foveated field of view area. As a result, this method achieves the 1.25 times super-resolution magnification in the foveated field of view area, which realizes the super-resolutio n imaging of local field while keeping the resolution of other fields unchanged. Experiments verify the feasibility and effectiveness of our system.

Utilization of ultrasound and glycation to improve functional properties and encapsulated efficiency of proteins in anthocyanins.

The purpose of this study is to explore the optimal conditions for the preparation of bovine serum albumin (BSA)/casein (CA)-dextran (DEX) conjugates by ultrasonic pretreatment combined with glycation (U-G treatment). When BSA and CA were treated with ultrasound (40% amplitude, 10 min), the grafting degree increased 10.57% and 6.05%, respectively. Structural analysis revealed that ultrasonic pretreatment changed the secondary structure, further affected functional properties of proteins. After U-G treatment, the solubility and thermal stability of BSA and CA was significantly increased, and the foaming and emulsifying capacity of proteins were also changed. Moreover, ultrasonic pretreatment and glycation exhibited a greater impact on BSA characterized with highly helical structure. Complexes fabricated by U-G-BSA/CA and carboxymethyl cellulose (CMC) exhibited protection on anthocyanins (ACNs), delaying the thermal degradation of ACNs. In conclusion, the protein conjugates treated by ultrasonic pretreatment combined with glycation have excellent functionality and are potential carrier materials.

A Novel Electrocatalyst Pd(II)@Ni3 (HITP)2 for Ultrasensitive Detection of Chloramphenicol: Experimental and Computational Investigation.

Ultrasensitive electrochemical detection on hazardous substances like antibiotics and pesticides is essential but still challenging in rapid test technology. Herein, the first electrode using highly conductive metal-organic frameworks (HCMOFs) for electrochemical detection of chloramphenicol is proposed. The design of electrocatalyst Pd(II)@Ni3 (HITP)2 with ultra-sensitivity in detection of chloramphenicol is demonstrated by loading Pd onto HCMOFs. An ultra-low limit of detection (LOD) of 0.2 nM (64.6 pg/mL) was observed for these materials, which is 1-2 orders of magnitude lower than the other reported materials for chromatographic detection. Moreover, the proposed HCMOFs showed long-time stability over 24 h. The superior detection sensitivity can be attributed to the high conductivity of Ni3 (HITP)2 , and the large Pd loading. The experimental characterizations and computational investigation determined the Pd loading mechanism in Pd(II)@Ni3 (HITP)2 , revealing PdCl2 adsorption on the abundant adsorption sites of Ni3 (HITP)2 . The proposed electrochemical sensor design using HCMOFs was demonstrated to be both effective and efficient, showing that the adoption of HCMOFs decorated with other effective electrocatalysts, which combine high conductivity and high catalytic activity, is of great advantage for ultrasensitive detection.

Image-based motion artifact reduction on liver dynamic contrast enhanced MRI.

Liver MRI images often suffer from degraded quality due to ghosting or blurring artifacts caused by patient respiratory or bulk motion. In this study, we developed a two-stage deep learning model to reduce motion artifact on dynamic contrast enhanced (DCE) liver MRIs. The stage-I network utilized a deep residual network with a densely connected multi-resolution block (DRN-DCMB) network to remove most motion artifacts. The stage-II network applied the generative adversarial network (GAN) and perceptual loss compensation to preserve image structural features. The stage-I network served as the generator of GAN and its pretrained parameters in stage-I were further updated via backpropagation during stage-II training. The stage-I network was trained using small image patches with simulated motion artifacts including image-space rotational and translational motion, and K-space based centric and interleaved linear motion, sinusoidal, and rotational motion to mimic liver motion patterns. The stage-II network training used full-size images with the same types of simulated motion. The liver DCE-MRI image volumes without obvious motion artifacts in 10 patients were used for the training process, of which 1020 images of 8 patients were used for training and 240 images of 2 patients for validation. Finally, the whole two-stage deep learning model was tested with simulated motion images (312 clean images from 5 test patients) and patient images with real motion artifacts (28 motion images from 12 patients). The resulted images after two-stage processing demonstrated reduced motion artifacts while preserved anatomic details without image blurriness, with SSIM of 0.935 ± 0.092, MSE of 60.7 ± 9.0 × 10-3, and PSNR of 32.054 ± 2.219.

Corrigendum: Network pharmacology-based prediction of active compounds in Wenyang Jiedu Huayu formula acting on acute-on-chronic liver failure with experimental support in vitro and in vivo.

[This corrects the article DOI: 10.3389/fphar.2022.1003479.].

Network pharmacology-based prediction of active compounds in the Wenyang Jiedu Huayu formula acting on acute-on-chronic liver failure with experimental support in vitro and in vivo.

Acute-on-chronic liver failure (ACLF) is characterized by undermined liver function, massive necrosis/apoptosis of hepatocytes, and hepatic inflammatory cell recruitment, leading to multiorgan failure. Traditional Chinese medicine (TCM) has been widely applied in clinical and experimental studies of ACLF. In this study, 23 compounds with 6,386 drug targets were obtained from Wenyang Jiedu Huayu (WYJDHY), and 8,096 genes were identified as ACLF disease targets, among which 3,132 were overlapping co-targets. Expression profile analysis identified 105 DEGs among the co-targets, which were associated with biological activities such as lymphocyte activation, immune response regulation, and pathways such as Th17 cell differentiation and NF-κB signaling. After PPI analysis and network construction, atractylenolide I (AT-1) has been identified as the hub active ingredient of the WYJDHY formula. LPS stimulation inhibited rat hepatocytes' BRL 3A cell viability, promoted cell apoptosis, increased the levels of ALT, AST, IL-6, and VCAM-1 within the culture medium, and activated NF-κB signaling, whereas AT-1 treatment significantly attenuated LPS-induced toxicity on BRL 3A cells. Furthermore, the NF-κB signaling inhibitor PDTC exerted effects on LPS-stimulated BRL 3A cells similar to those of AT-1, and the combination of PDTC and AT-1 further attenuated LPS-induced toxicity on BRL 3A cells. In vivo, AT-1 alone or with PDTC improved the symptoms and local inflammation in ACLF model rats. In conclusion, 23 active ingredients of six herbs in the WYJDHY formula were retrieved, and 105 co-targets were differentially expressed in ACLF. AT-1 exerts protective effects on LPS-stimulated hepatocytes and ACLF rats, possibly by inhibiting the NF-κB pathway.

A multicohort geometric deep learning study of age dependent cortical and subcortical morphologic interactions for fluid intelligence prediction.

The relationship of human brain structure to cognitive function is complex, and how this relationship differs between childhood and adulthood is poorly understood. One strong hypothesis suggests the cognitive function of Fluid Intelligence (Gf) is dependent on prefrontal cortex and parietal cortex. In this work, we developed a novel graph convolutional neural networks (gCNNs) for the analysis of localized anatomic shape and prediction of Gf. Morphologic information of the cortical ribbons and subcortical structures was extracted from T1-weighted MRIs within two independent cohorts, the Adolescent Brain Cognitive Development Study (ABCD; age: 9.93 ± 0.62 years) of children and the Human Connectome Project (HCP; age: 28.81 ± 3.70 years). Prediction combining cortical and subcortical surfaces together yielded the highest accuracy of Gf for both ABCD (R = 0.314) and HCP datasets (R = 0.454), outperforming the state-of-the-art prediction of Gf from any other brain measures in the literature. Across both datasets, the morphology of the amygdala, hippocampus, and nucleus accumbens, along with temporal, parietal and cingulate cortex consistently drove the prediction of Gf, suggesting a significant reframing of the relationship between brain morphology and Gf to include systems involved with reward/aversion processing, judgment and decision-making, motivation, and emotion.

Phonetic category activation predicts the direction and magnitude of perceptual adaptation to accented speech.

Unfamiliar accents can systematically shift speech acoustics away from community norms and reduce comprehension. Yet, limited exposure improves comprehension. This perceptual adaptation indicates that the mapping from acoustics to speech representations is dynamic, rather than fixed. But, what drives adjustments is debated. Supervised learning accounts posit that activation of an internal speech representation via disambiguating information generates predictions about patterns of speech input typically associated with the representation. When actual input mismatches predictions, the mapping is adjusted. We tested two hypotheses of this account across consonants and vowels as listeners categorized speech conveying an English-like acoustic regularity or an artificial accent. Across conditions, signal manipulations impacted which of two acoustic dimensions best conveyed category identity, and predicted which dimension would exhibit the effects of perceptual adaptation. Moreover, the strength of phonetic category activation, as estimated by categorization responses reliant on the dominant acoustic dimension, predicted the magnitude of adaptation observed across listeners. The results align with predictions of supervised learning accounts, suggesting that perceptual adaptation arises from speech category activation, corresponding predictions about the patterns of acoustic input that align with the category, and adjustments in subsequent speech perception when input mismatches these expectations. (PsycInfo Database Record (c) 2022 APA, all rights reserved).

Deep Gaussian processes for multiple instance learning: Application to CT intracranial hemorrhage detection.

BACKGROUND AND OBJECTIVE: Intracranial hemorrhage (ICH) is a life-threatening emergency that can lead to brain damage or death, with high rates of mortality and morbidity. The fast and accurate detection of ICH is important for the patient to get an early and efficient treatment. To improve this diagnostic process, the application of Deep Learning (DL) models on head CT scans is an active area of research. Although promising results have been obtained, many of the proposed models require slice-level annotations by radiologists, which are costly and time-consuming. METHODS: We formulate the ICH detection as a problem of Multiple Instance Learning (MIL) that allows training with only scan-level annotations. We develop a new probabilistic method based on Deep Gaussian Processes (DGP) that is able to train with this MIL setting and accurately predict ICH at both slice- and scan-level. The proposed DGPMIL model is able to capture complex feature relations by using multiple Gaussian Process (GP) layers, as we show experimentally. RESULTS: To highlight the advantages of DGPMIL in a general MIL setting, we first conduct several controlled experiments on the MNIST dataset. We show that multiple GP layers outperform one-layer GP models, especially for complex feature distributions. For ICH detection experiments, we use two public brain CT datasets (RSNA and CQ500). We first train a Convolutional Neural Network (CNN) with an attention mechanism to extract the image features, which are fed into our DGPMIL model to perform the final predictions. The results show that DGPMIL model outperforms VGPMIL as well as the attention-based CNN for MIL and other state-of-the-art methods for this problem. The best performing DGPMIL model reaches an AUC-ROC of 0.957 (resp. 0.909) and an AUC-PR of 0.961 (resp. 0.889) on the RSNA (resp. CQ500) dataset. CONCLUSION: The competitive performance at slice- and scan-level shows that DGPMIL model provides an accurate diagnosis on slices without the need for slice-level annotations by radiologists during training. As MIL is a common problem setting, our model can be applied to a broader range of other tasks, especially in medical image classification, where it can help the diagnostic process.