Deep learning improves acoustic biodiversity monitoring and new candidate forest frog species identification (genus Platymantis) in the Philippines

Purpose: Although deep learning (DL) models have shown increasing ability to accurately classify diagnostic images in oncology, significantly large amounts of well-curated data are often needed to match human level performance. Given the relative paucity of imaging datasets for less prevalent cancer types, there is an increasing need for methods which can improve the performance of deep learning models trained using limited diagnostic images. Deep metric learning (DML) is a potential method which can improve accuracy in deep learning models trained on limited datasets. Leveraging a triplet-loss function, DML exponentially increases training data compared to a traditional DL model. In this study, we investigated the utility of DML to improve the accuracy of DL models trained to classify cancerous lesions found on screening mammograms. Methods: Using a dataset of 2620 lesions found on routine screening mammogram, we trained both a traditional DL and DML models to classify suspicious lesions as cancerous or benign. The VGG16 architecture was used as the basis for the DL and DML models. Model performance was compared by calculating model accuracy, sensitivity, and specificity on a blinded test set of 378 lesions. In addition to individual model performance, we also measured agreement accuracy when both the DL and DML models were combined. Sub-analyses were conducted to identify phenotypes which were best suited for each model type. Both models underwent hyperparameters optimization to identify ideal batch size, learning rate, and regularization to prevent overfitting. Results: We found that the combination of the traditional DL model with DML model resulted in the highest overall accuracy (78.7%) representing a 7.1% improvement compared to the traditional DL model (p<.001). Alone, the traditional DL model had an improved accuracy compared to the DML model (71.4% vs 66.4%). The traditional DL model had a higher sensitivity (94.8% vs 73.6 %) , but lower specificity (34.7% vs 55.1%) compared the DML model. Sub-analyses suggested the traditional DL model was more accurate on higher density breasts, whereas the DML model was more accurate on lower density breasts. Additionally, the traditional DL model had the highest accuracy on oval shaped lesions, compared to the DML model which was most accurate on irregularly shaped breast lesions. Conclusion: Our study suggests that addition of DML models with traditional DL models can improve diagnostic image classification performance in cancer. Our results suggest DML models may provide increased specificity and help with classification of unique populations often misclassified by traditional DL models. Further studied investigating the utility of DML on other cancer imaging tasks are necessary to successfully build more robust DL models in cancer imaging. Citation Format: Justin Du, Sachin Umrao, Enoch Chang, Marina Joel, Aidan Gilson, Guneet Janda, Rachel Choi, Yongfeng Hui, Sanjay Aneja. The utility of deep metric learning for breast cancer identification on mammographic images [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2021; 2021 Apr 10-15 and May 17-21. Philadelphia (PA): AACR; Cancer Res 2021;81(13_Suppl):Abstract nr 184.

Deep learning has become increasingly popular due to its remarkable ability to learn high-dimensional feature representations. Numerous algorithms and models have been developed to enhance the application of deep learning across various real-world tasks, including image classification, natural language processing, and autonomous driving. However, deep learning models are susceptible to backdoor threats, where an attacker manipulates the training process or data to cause incorrect predictions on malicious samples containing specific triggers, while maintaining normal performance on benign samples. With the advancement of deep learning, including evolving training schemes and the need for large-scale training data, new threats in the backdoor domain continue to emerge. Conversely, backdoors can also be leveraged to protect deep learning models, such as through watermarking techniques. In this thesis, we conduct an in-depth investigation into backdoor techniques from three novel perspectives. In the first part of this thesis, we demonstrate that emerging deep learning training schemes can introduce new backdoor risks. Specifically, pre-trained Natural Language Processing (NLP) models can be easily adapted to a variety of downstream language tasks, significantly accelerating the development of language models. However, the pre-trained model becomes a single point of failure for these downstream models. We propose a novel task-agnostic backdoor attack against pre-trained NLP models, wherein the adversary does not need prior information about the downstream tasks when implanting the backdoor into the pre-trained model. Any downstream models transferred from this malicious model will inherit the backdoor, even after extensive transfer learning, revealing the severe vulnerability of pre-trained foundation models to backdoor attacks. In the second part of this thesis, we develop novel backdoor attack methods suited to new threat scenarios. The rapid expansion of deep learning models necessitates large-scale training data, much of which is unlabeled and outsourced to third parties for annotation. To ensure data security, most datasets are read-only for training samples, preventing the addition of input triggers. Consequently, attackers can only achieve data poisoning by uploading malicious annotations. In this practical scenario, all existing data poisoning methods that add triggers to the input are infeasible. Therefore, we propose new backdoor attack methods that involve poisoning only the labels without modifying any input samples. In the third part of this thesis, we utilize the backdoor technique to proactively protect our deep learning models, specifically for intellectual property protection. Considering the complexity of deep learning tasks, generating a well-trained deep learning model requires substantial computational resources, training data, and expertise. Therefore, it is essential to protect these assets and prevent copyright infringement. Inspired by backdoor attacks that can induce specific behaviors in target models through carefully designed samples, several watermarking methods have been proposed to protect the intellectual property of deep learning models. Model owners can train their models to produce unique outputs for certain crafted samples and use these samples for ownership verification. While various extraction techniques have been designed for supervised deep learning models, challenges arise when applying them to deep reinforcement learning models due to differences in model features and scenarios. Therefore, we propose a novel watermarking scheme to protect deep reinforcement learning models from unauthorized distribution. Instead of using spatial watermarks as in conventional deep learning models, we design temporal watermarks that minimize potential impact and damage to the protected deep reinforcement learning model while achieving high-fidelity ownership verification. In summary, this thesis investigates the evolving landscape of backdoor threats during the development of deep learning techniques and the use of backdoors for beneficial purposes in intellectual property protection.

We describe a new species of high-elevation rainforest tree frog (genus Platymantis) from the Nakanai Mountains, New Britain Island, Bismarck Archipelago, Southwestern Pacific. The new species is characterized by moderate body size (34.2–35.8 mm for four males), widely expanded terminal digital disks of the fingers and toes, smooth skin of the dorsum, a distinctive color pattern, and a complex, amplitude-modulated advertisement call produced in groups of 3–6 notes. We compare the new species to all known species of Platymantis from New Britain and to additional phenotypically similar species from the Solomon Islands and Fiji. It is most similar to P. nexipus, a species known previously from only a single specimen. We rediagnose and redescribe P. nexipus on the basis of the holotype and ten recently collected specimens, provide the first descriptions of the advertisement calls of both species, and comment on an additional suspected undescribed species from the Nakanai Mountains of New Britain Island.

There are over 9000 bird and frog species in the globe. Some of the species are rare to find, and even when they are, predicting their behaviour is challenging. There is an efficient and simple technique to recognise these frog and bird species contingent on their traits to solve this challenge. Also, humans are better at recognising birds and frogs through sounds than they are at recognising them through photographs. As a result, employed various CNN models including CNN-Sequential, CNN-ResNet, CNN-EfficientNet, CNN-VGG19 and a hybrid model Convolution Neural Networks with Long Short-term Memory (CNN-LSTM). It is a powerful deep learning model that has shown to be effective in image processing. Compared to standard alone models, hybrid model produces better accuracy. A hybrid system for classifying bird and frog species is provided in this study, which employs the Rainforest Connection Species Audio Detection dataset from Kaggle repository for both training and testing. The classification of bird or frog species by using audio dataset after processing it and convert it into spectrogram images. Among all the deployed models CNN-LSTM system has been shown to achieve satisfactory results in practise by building this dataset and achieves accuracy of 92.47.

We describe two new species of forest frogs in the genus Platymantis from the Isabel Island group, Solomon Islands. One new species is a medium-sized, terrestrial form that is morphologically most similar to P. weberi (a widespread Solomon Islands species). The other new species is an arboreal frog that is morphologically similar to Platymantis neckeri (known from Bougainville, Choiseul, and Isabel islands). Both new species possess unique advertisement calls that distinguish them from all sympatric congeners. Because acoustic characteristics function as the primary mate-recognition signals for anuran species, and are therefore an excellent indicator of the status of unique evolutionary lineages, we recognize each as new species. We diagnose both new species on the basis of their distinctive advertisement calls and in the case of the terrestrial form, by differences in body size, body proportions and skin texture. The diversity of ceratobatrachid frogs of the Solomon islands and Bougainville is underestimated and in need of a comprehensive taxonomic review coupled with a standardized survey of acoustic characters.

The need for intubation in methanol-poisoned patients, if not predicted in time, can lead to irreparable complications and even death. Artificial intelligence (AI) techniques like machine learning (ML) and deep learning (DL) greatly aid in accurately predicting intubation needs for methanol-poisoned patients. So, our study aims to assess Explainable Artificial Intelligence (XAI) for predicting intubation necessity in methanol-poisoned patients, comparing deep learning and machine learning models. This study analyzed a dataset of 897 patient records from Loghman Hakim Hospital in Tehran, Iran, encompassing cases of methanol poisoning, including those requiring intubation (202 cases) and those not requiring it (695 cases). Eight established ML (SVM, XGB, DT, RF) and DL (DNN, FNN, LSTM, CNN) models were used. Techniques such as tenfold cross-validation and hyperparameter tuning were applied to prevent overfitting. The study also focused on interpretability through SHAP and LIME methods. Model performance was evaluated based on accuracy, specificity, sensitivity, F1-score, and ROC curve metrics. Among DL models, LSTM showed superior performance in accuracy (94.0%), sensitivity (99.0%), specificity (94.0%), and F1-score (97.0%). CNN led in ROC with 78.0%. For ML models, RF excelled in accuracy (97.0%) and specificity (100%), followed by XGB with sensitivity (99.37%), F1-score (98.27%), and ROC (96.08%). Overall, RF and XGB outperformed other models, with accuracy (97.0%) and specificity (100%) for RF, and sensitivity (99.37%), F1-score (98.27%), and ROC (96.08%) for XGB. ML models surpassed DL models across all metrics, with accuracies from 93.0% to 97.0% for DL and 93.0% to 99.0% for ML. Sensitivities ranged from 98.0% to 99.37% for DL and 93.0% to 99.0% for ML. DL models achieved specificities from 78.0% to 94.0%, while ML models ranged from 93.0% to 100%. F1-scores for DL were between 93.0% and 97.0%, and for ML between 96.0% and 98.27%. DL models scored ROC between 68.0% and 78.0%, while ML models ranged from 84.0% to 96.08%. Key features for predicting intubation necessity include GCS at admission, ICU admission, age, longer folic acid therapy duration, elevated BUN and AST levels, VBG_HCO3 at initial record, and hemodialysis presence. This study as the showcases XAI's effectiveness in predicting intubation necessity in methanol-poisoned patients. ML models, particularly RF and XGB, outperform DL counterparts, underscoring their potential for clinical decision-making.

We describe a new species of forest frog in the genus Platymantis from New Britain Island, Bismark Archipelago, Papua New Guinea. The new species is a morphologically cryptic form that has masqueraded for almost four decades under the name P. schmidti (formerly P. papuensis schmidti, Brown & Tyler, 1968). The new species is microsympatric with the geographically widespread P. schmidti at two known localities. We diagnose the new species on the basis of its distinctive advertisement call and slight but consistent differences in body size and proportions. Calling males of the new species appear to prefer more elevated perches than do males of P. schmidti and the new species may exhibit a greater extent of sexual size dimorphism.

Male calls play different roles in anuran social organization, such as spacing, territoriality and female attraction. However, calls and associated behaviors remain poorly described for many anuran species. Here we describe the advertisement and release calls of the tree frog Phyllomedusa ayeaye (Lutz, 1966) and report on the social context of emissions and a physical combat. Approximately 35 minutes of digital recordings were obtained from 34 hours of observations at one breeding site in the state of Minas Gerais, southeastern Brazil. Bioacoustic analysis showed that males emitted two types of advertisement calls: 1) simple call (a sequence of short pulsed notes) and 2) composite call (a sequence of short pulsed notes followed by a long pulsed note). Composite calls were emitted more frequently during more intense chorus activity, with various active males at the breeding site. The release call was also composed by short pulsed notes, with a wider spectrum of frequencies and emitted more rapidly than the advertisement calls. Our results suggest that the composite call of P. ayeaye may represent a mixed advertisement call. Long notes might be the aggressive part directed to males, whereas short notes directed to females. Our description of call types, their functions, and male physical interactions will be useful for studies investigating the systematics and behavior of Phyllomedusa species.

The use of acoustic signals is a common characteristic of most anuran species to mediate intraspecific communication. Besides many social purposes, one of the main functions of these signals is species recognition. For this reason, this phenotypic trait is normally applied to taxonomy or to construct evolutionary relationship hypotheses. Here the acoustic repertoire of five populations of the genus Allobates from the Brazilian Atlantic Forest are presented for the first time, on a vulnerable to extinction Neotropical taxon. The description of males’ advertisement and aggressive calls and a female call emitted in a courtship context are presented. In addition, the advertisement calls of individuals from distinct geographical regions were compared. Differences in frequency range and note duration may imply in taxonomic rearrangements of these populations, once considered distinct species, and more recently, proposed as a single species, Allobatesolfersioides. Calls of the male from the state of Rio de Janeiro do not overlap spectrally with calls of males from northern populations, while the shorter notes emitted by males from Alagoas also distinguishes this population from the remaining southern populations. Therefore, it is likely that at least two of the junior synonyms should be revalidated. Similarities among male advertisement and female calls are generally reported in other anuran species; these calls may have evolved from a preexisting vocalization common to both sexes. Male aggressive calls were different from both the male advertisement and female calls, since it was composed by a longer and multi-pulsed note. Aggressive and advertisement calls generally have similar dominant frequencies, but they have temporal distinctions. Such patterns were corroborated with the Atlantic Forest Rocket Frogs. These findings may support future research addressing the taxonomy of the group, behavioral evolution, and amphibian conservation.

The classification and identification of individual tree species in forest environments are critical for biodiversity conservation, sustainable forestry management, and ecological monitoring. Recent advances in drone technology and artificial intelligence have enabled new methodologies for detecting and classifying trees at an individual level. However, significant challenges persist, particularly in heterogeneous forest environments with high species diversity and complex canopy structures. This systematic review explores the latest research on drone-based data collection and AI-driven classification techniques, focusing on studies that classify specific tree species rather than generic tree detection. Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines, peer review studies from the last decade were analyzed to identify trends in data acquisition instruments (e.g., RGB, multispectral, hyperspectral, LiDAR), preprocessing techniques, segmentation approaches, and machine learning (ML) algorithms used for classification. Findings of this study reveal that deep learning (DL) models, particularly convolutional neural networks (CNN), are increasingly replacing traditional ML methods such as random forest (RF) or support vector machines (SVMs) because there is no need for a feature extraction phase, as this is implicit in the DL models. The integration of LiDAR with hyperspectral imaging further enhances classification accuracy but remains limited due to cost constraints. Additionally, we discuss the challenges of model generalization across different forest ecosystems and propose future research directions, including the development of standardized datasets and improved model architectures for robust tree species classification. This review provides a comprehensive synthesis of existing methodologies, highlighting both advancements and persistent gaps in AI-driven forest monitoring.

Study question Can heatmaps generated by occlusion explain the patterns learned by deep learning (DL) models classifying the embryo viability in ART? Summary answer Occlusion experiments generate heatmaps that reveal which regions in frames of time-lapse video (TLV) are more discriminative for classification and prediction by the DL models. What is known already DL has widely been explored in ART for embryo selection. Depending upon input (video or image), different DL models classifying embryo viability are developed. However, whether the prediction is based on actual input features or random guessing is unknown. The embryo selection in ART is subjective. If the intention is using DL models’ prediction to transfer, freeze or discard the embryo, explanations of how they interpret embryonic development features brings transparency and trust. In other areas, heatmaps are used for explaining DL predictions. The heatmaps can be a tool to understand patterns learned by DL models for embryo selection. Study design, size, duration We trained two separate DL models for predicting the presence of fetal heartbeat for the transferred embryos. We further used occlusion generated heatmaps to explain the predictions. For training, retrospective data was used. The input dataset consisted of 136 TLVs and corresponding patient data for 132 participants (128: single embryo transfers and 8: double embryo transfer) from both IVF and ICSI treatment. Each video was assessed by an embryologist. Participants/materials, setting, methods DL models (A as ResNet–18, B as VGG16) are trained for predicting the presence of fetal heartbeat on a single frame extracted from TLV after day three or later. Model A has a better recall (0.7) compared to B (0.5). Heatmaps explain the reason behind models’ recall rate by visually representing patterns learned by them. Using occlusion filter size 30*30 with stride 14 and size 50*50 with stride 25, we generate heatmaps for both models. Main results and the role of chance The heatmaps generated using occlusion can represent visually the patterns discovered by the DL models when predicting the presence of a fetal heartbeat. Using occlusion filter size 30*30 with stride 14, we verified that Model B has lower recall because the heatmaps show that the model finds redundant features present outside the embryo region in many input frames. It could be interpreted that either the model has not learned relevant patterns or is more robust to noise. This representation of DL models equips us in better decision-making, whether to consider or discard the prediction or rather train the model further, preprocess training data or change network architecture. The heatmaps revealed that for frames where significant patterns learned by the models are within the embryo region, more weight was given to specific features like the inner cell mass, trophectoderm and some parts within the zona pellucida. Moreover, the heat maps generated using occlusion are independent of the underlying model’s architecture as the same experiment settings were used for both models. For occlusion filter size 50*50 with stride 25, the expanse of input regions (in or outside the embryo) considered relevant could be visualized for both models A and B. Limitations, reasons for caution Heatmaps generated by occluding input regions give a visual representation of features in individual frames not directly on videos. Explaining DL models by heatmaps besides occlusion, other techniques (Grad-Cam) exist but were not evaluated. Furthermore, there is no quantitative measure for evaluating whether heatmaps are a good explanation or not. Wider implications of the findings: The heatmaps make the patterns discovered by DL models visually recognized and bring forth the prominent portions of embryo regions. This will again improve understanding and trust in DL models’ predictions. Visual representation of DL models using heatmaps enables interpreting a prediction, performing model analysis and determining scope for improvement. Trial registration number Not applicable

-This study investigates predator size and prey type as potential proximal causes of differences among diets of three lizard species (family Scincidae) and three frog species (subfamily Limnodynastinae) that coexist in wet subtropical forest in eastern Australia. Frogs eat smaller prey than lizards having the same gape size and there were significant differences in the types of arthropods eaten by frogs and lizards. Differences among species within frogs and lizards were small and not statistically significant Frogs ate more amphipods, mites, and ants than the lizards, and lizards ate more termites, millipedes, isopods, and orthopterans than the frogs. Other categories were eaten in similar quantities by both frogs and lizards. The degree of specialization in types and sizes of prey often changes with the body size of a predator. Change of diet with ontogeny has been related to changes in prey size in lizards (Schoener and Gorman, 1968; Rose, 1976; Dominguez and Salvador, 1990; Magnusson and Silva, 1993). Frogs change both prey type and prey size as they grow (Pengilley, 1971; Labanick, 1976; Christian, 1982; Donnelly, 1991; Simon and Toft, 1991; Wiggins, 1992). The latter authors suggested that the change in prey types is a result of the shift in prey size, because different types of arthropods have different mean sizes. However, diet composition differs among species in some assemblages of frogs (Lima and Magnusson, 1998) and lizards (Magnusson and Silva, 1993), and the shift in prey types with growth is more than a passive effect of selection for larger prey in seven species of leaf-litter frogs of Central Amaz6nia (Lima and Moreira, 1993; Lima, 1998). Caldwell and Vitt (1999) showed consistent differences between species of lizards and species of frogs in one Amazonian locality, but there are no other published studies of differences in diet between syntopic lizards and frogs. In this study, we make use of extensive collections of subtropical lizards and frogs in the Australian Museum to investigate the effects of predator size and species identity on diet composition within and between three species of lizards (family Scincidae) and three species of frogs (subfamily Limnodynastinae) that coexist in subtropical rainforest in eastern Australia. MATERIALS AND METHODS The frogs and lizards were collected during a New South Wales National Parks and Wildlife 40 This content downloaded from 157.55.39.239 on Thu, 15 Sep 2016 05:29:31 UTC All use subject to http://about.jstor.org/terms DIFFERENCES IN DIET AMONG FROGS AND LIZARDS Service (NSW NPWS) project coordinated by Harry Hines between 1988 and 1992 and deposited in the Australian Museum in Sydney. The collections were made in Nightcap National Park, Dome Mountain Area, Eastern Border Ranges National Park, Mount Warning National Park, Spirabo State Forest, Forestland State Forest, and adjacent areas in areas of humid subtropical forest in northern New South Wales. In most areas, pitfall traps with formalin were used by members of the NSW NPWS survey team to sample invertebrates, and the capture of vertebrates was accidental. Frogs and lizards (35% of individuals) deposited in the museum that had been collected in the same region but lacking exact geographic coordinates were used to increase sample sizes. We used only specimens that were available in the collection of the Australian museum and did not collect or kill any of the animals ourselves. Only three species of scincid lizards, Calyptotis scutirostrum (N = 39), Eulamprus murrayi (N = 23), and Saproscincus challengeri (N = 30) and three species of frogs of the family Myobatrachidae (Limnodynastinae), Assa darlingtoni (N = 36), Lechriodus fletcheri (N = 26), and Philoria loveridgei (N = 18) were sufficiently common and had enough items in the stomach to justify analysis. The animals were fixed in formalin and maintained in

Modern frogs and toads possess a structurally unique saccule, endowing them with seismic sensitivity greater than that observed so far in any other group of terrestrial vertebrates. In synchrony with their advertisement calls, approximately half of the calling males of one frog species, the Puerto-Rican white-lipped frog (Leptodactylus albilabris), produce impulsive seismic signals (thumps). The spectral distribution of power in these seismic signals matches precisely the spectral sensitivity of the frog's saccule. The signals have sufficient amplitude to be sensed easily by the frog's saccule up to several meters from the source—well beyond the typical spacing when these frogs are calling in a group. This circumstantial evidence suggests that white-lipped frogs may use the seismic channel in intraspecific communication, possibly as an alternative to the airborne channel, which often is cluttered with noise and interference. Using the frog's vocalizations as our assay, we set out to test that proposition. In response to playback calls, the male white-lipped frog adjusts several of its own calling parameters. The most conspicuous of these involves call timing—specifically the tendency for a gap in the distribution of call onsets, precisely timed with respect to the onsets of the playback calls. When the airborne component is unavailable (e.g., masked by noise), approximately one in five animals produces the calling gap in response to the seismic signals alone.

ABSTRACTEvolutionary divergence in sexual signals may lead to or maintain reproductive isolation between populations. Both selective forces—such as ecological and sexual selection—and random processes like genetic drift may influence the diversification of sexual signals. Understanding the patterns and sources of intraspecific variation in sexual signals among populations can inform the stages of differentiation and speciation. In this study, we investigated patterns of geographic variation in vocal sexual signals and how they relate to genetic and environmental distances among nine populations of Pacific tree frogs. Importantly, the taxonomy of Pacific tree frogs remains unresolved; while some authors recognize only one species (Pseudacris regilla), other authors propose three distinct species based on mitochondrial DNA lineages (P. regilla, P. sierra, and P. hypochondriaca). Our genetic analyses revealed that the nine populations studied span two of the three mitochondrial lineages of Pacific tree frogs. We found that variation in the advertisement calls is better explained by mitochondrial lineage than by geographic or environmental distances between populations. The acoustic properties that have diverged the most between lineages relate to the number of pulses in the call and the rate at which the pulses are delivered. Interestingly, these acoustic properties are important for species recognition in this and other species of frogs. These findings suggest that differences in the vocal sexual signal may lead to premating reproductive isolation between mitochondrial lineages of Pacific tree frogs.

ObjectiveMicrovascular invasion (MVI) is of great significance for the individualized treatment of hepatocellular carcinoma (HCC) and preoperative noninvasive prediction of MVI is still an urgent clinical problem. To explore the effects of different regions of interest (ROI) and image input dimensions on the performance of deep learning (DL) models, and to select the best result to develop and validate a DL model for preoperative prediction of MVI.Materials and methodsA total of 206 patients with pathologically confirmed HCC from three hospitals were retrospectively enrolled and divided into training, internal validation and external test set. Based on hepatobiliary phase images (HBP) of gadoxetic acid-enhanced MRI, 2D DL, 3D DL and 2.5D deep multi-instance learning (MIL) models were established. The receiver operating characteristic curve (ROC) was used to evaluate the predictive efficacy of the above models. Based on the optimal performance model, the T1WI-FS and T2WI-FS images were preprocessed correspondingly, and a multimodal prediction model including three sequences was constructed. The ROC, and decision curve were used to visualize the predictive ability of the model.ResultsCompared with 2D DL and 3D DL models, the 2.5D DL model based on all axial images of ROI had the highest performance, with the AUC values of 0.802 (95% CI, 0.669–0.936) and 0.759 (95% CI, 0.643–0.875) in the validation and test sets. The AUCs of the multimodal MRI model were 0.954 (95% CI, 0.920–0.989) in the training set, 0.857 (95% CI, 0.736–0.978) in the validation set, and 0.788 (95% CI, 0.681–0.895) in the test set.ConclusionThe DL model that selects all axial slices of intratumor and peritumor as input shows robust capability in predicting MVI, which is expected to help clinical decision-making of individualized treatment for HCC.Supplementary InformationThe online version contains supplementary material available at 10.1186/s12885-025-14971-7.