The advent of Large Language Models (LLMs) and other advances in Artificial Intelligence (AI) has revolutionized natural language processing across a wide range of domains. However, applying these to high-stakes domains, such as medical question answering, remains challenging, particularly in ensuring factual accuracy and navigating complex, multi-step reasoning. We present TRUSS-Med, a system that utilizes specialized LLM agents to collaborate and automate the Medical Question Answering task by breaking it down into simpler, more manageable sub-tasks. We integrate the generative capabilities of LLMs with the structured reasoning of State Space Models (SSMs) to retrieve information, generate answers, and confirm their accuracy. Experiments on the MedQA-USMLE and MedMCQA datasets achieve strong performance, demonstrating notable improvements of 30.63 and 17.44 percentage points in accuracy over GPT-3.5. TRUSS-Med demonstrates notable performance gains over comparable models, as evidenced by results against GPT-3.5.

This paper presents EmoMamba, a novel framework designed for dynamic facial emotion recognition by using both visual and textual input. The proposed approach consists of a feature encoder stacked with our proposed GBMamba layers. Initially, a vision backbone is trained on a static facial emotion dataset to extract meaningful features. Subsequently, cropped facial frames are encoded using the vision backbone to produce sequences of facial features. These sequences are processed by an encoder that incorporates GBMamba layers. Moreover, the video description text is encoded using the Mamba module. The visual and textual features are then concatenated and passed through an EmoMamba multimodal encoder to effectively integrate both modalities. Experimental evaluations on dynamic facial emotion datasets, such as MAFW and DFEW, demonstrate the effectiveness and potential of the proposed method.

Neonatal jaundice is a common and potentially fatal health condition in neonates, especially in low and middle income countries, where it contributes considerably to neonatal morbidity and death. Traditional diagnostic approaches, such as Total Serum Bilirubin (TSB) testing, are invasive and could lead to discomfort, infection risk, and diagnostic delays. As a result, there is a rising interest in non-invasive approaches for detecting jaundice early and accurately. An in-depth analysis of non-invasive techniques for detecting neonatal jaundice is presented by this review, exploring several AI-driven techniques, such as Machine Learning (ML) and Deep Learning (DL), which have demonstrated the ability to enhance diagnostic accuracy by evaluating complex patterns in neonatal skin color and other relevant features. It is identified that AI models incorporating variants of neural networks achieve an accuracy rate of over 90% in detecting jaundice when compared to traditional methods. Furthermore, satisfactory outcomes in field settings have been demonstrated by mobile-based applications that use smartphone cameras to estimate bilirubin levels, providing a practical alternative for resource-constrained areas. The potential impact of AI-based solutions on reducing neonatal morbidity and mortality is evaluated by this review, with a focus on real-world clinical challenges, highlighting the effectiveness and practicality of AI-based strategies as an assistive tool in revolutionizing neonatal care through early jaundice diagnosis, while also addressing the ethical and practical implications of integrating these technologies in clinical practice. Future research areas, such as the development of new imaging technologies and the incorporation of wearable sensors for real-time bilirubin monitoring, are recommended by the paper. • AI-driven methods achieve over 90% accuracy in neonatal jaundice detection. • Mobile-based applications offer practical solutions in resource-limited settings. • Emphasis on potential of AI-based strategies in early neonatal jaundice detection. • Challenges include device calibration and accuracy variations across skin tones. • Future research on novel imaging technologies for real-time jaundice monitoring.

Single image super-resolution generation is a computer vision task that aims to enhance the detail and quality of a low-resolution image, generating a higher-resolution version. To date, several attempts have been made to approach this problem using convolutional neural networks and, more recently, GANs, diffusion models, and transformers. The single-image super-resolution task largely involves the recovery of missing high-frequency details. This has led to the exploration of methods based on the Fourier and wavelet transform. In this work we focused on transformer-based models and wavelets, with the intention of exploiting the advantages of both: the former for its expressive capacity and remarkable achievements in the state of the art, the latter for its ability to decompose a signal into informative components. The proposed model, Discrete Wavelet Transform Super Resolution Network (DWTSRNet), is a transformer-based model composed of three parts: a shallow feature extraction module utilizing the discrete wavelet transform, a deep feature extraction module composed of several residual blocks of SwinV2 Transformer, and a reconstruction module using the inverse discrete wavelet transform. We have conducted experiments on the super-resolution of medical images, specifically focusing on MRI slices of the human brain. The experimental results show that DWTSRNet outperformed other wavelet-based models on standard benchmark datasets and demonstrated promising results on medical datasets. Our code is available at https://github.com/pcestola/DWTSRNet.

Knee pain is the most common disabling symptom in osteoarthritis (OA). High correlation between knee pain and multiple OA features is reported in literature, but it has not been validated using deep learning models. In this study, we aim to develop a deep hybrid learning model for pain prediction directly from radiography images. We obtain an optimal hybrid model with VGG16, GAP, and KNN combination that gave a maximum of 89.75% accuracy and 0.91 Cohen’s kappa scores. The pain prediction of our proposed approach has achieved 0.99 of receiver operating characteristic area under curve (ROC-AUC). Binary pain classification has demonstrated better precision-recall curve pattern as compared to 11-class, 4-class, and 3-class pain prediction tasks. Based on Gradient-weighted Class Activation Mapping (GradCAM) analysis, joint center was identified as a key area that significantly contributes to the network’s decision-making process. The results of this study demonstrate the capability of hybrid deep learning model in predicting baseline pain severity from plain radiographs, therefore improving future OA pain assessment efforts.

Post-traumatic stress disorder (PTSD) is a mental disorder that can be developed after witnessing or experiencing extremely traumatic events. PTSD can affect anyone, regardless of ethnicity, or culture. An estimated one in every eleven people will experience PTSD during their lifetime. The Clinician-Administered PTSD Scale (CAPS) and the PTSD Check List for Civilians (PCL-C) interviews are gold standards in the diagnosis of PTSD. These questionnaires can be fooled by the subject's responses. This work proposes a deep learning-based approach that achieves state-of-the-art performances for PTSD detection using audio recordings during clinical interviews. Our approach is based on MFCC low-level features extracted from audio recordings of clinical interviews, followed by deep high-level learning using a Stochastic Transformer. Our proposed approach achieves state-of-the-art performances with an RMSE of 2.92 on the eDAIC dataset thanks to the stochastic depth, stochastic deep learning layers, and stochastic activation function.

Parkinson's disease is a neurodegenerative disorder that affects movement and muscle control and is caused by the loss of dopamine-producing neurons in the brain. The main symptoms of Parkinson's disease (PD) include tremors, rigidity, slowness of movement, imbalances, and linguistic impairment. One of the most pronounced clinical indicators is a change in the patient's voice, which can be used to assist in the diagnosis and evaluation of PD. An innovative method based on speech signals is proposed in this study to automatically identify PD by a sophisticated learning strategy to extract features via a parallel convolution-based network with an attention mechanism to preferentially focused on relevant PD cues. The proposed method utilized raw speech and i-vector as input tensors. We evaluated the method by different metrics including accuracy 98%, precision 0.99, recall 0.96, and f1-score 0.97 which shows the model's robustness.

Depression is a serious mental disorder that affects many individuals across the globe. Depression (unipolar or bipolar) is characterized by a high rate of relapse or recurrence where a person might experience depressive episodes after non-depressive ones. The symptom patterns for recurrent depressive episodes have not been properly analyzed. Thus, there is a pressing need for systems which can monitor the mental health of individuals at risk to detect initial signs of relapse and recurrence. This points towards an automated system which identifies such signs and facilitates in timely treatment. In this paper, we introduce for the first time a deep learning based prospective monitoring system for the identification of relapse signs using audio-visual cues. The proposed model approximates relapse as the similarity between non-depression and depression samples. Experiments were performed on the DAIC-WOZ dataset and a highest accuracy of 73.21% was obtained using a Siamese network-based approach with one-shot learning regime.

The human face conveys a significant amount of information. Through facial expressions, the face is able to communicate numerous sentiments without the need for verbalisation. Visual emotion recognition has been extensively studied. Recently several end-to-end trained deep neural networks have been proposed for this task. However, such models often lack generalisation ability across datasets. In this paper, we propose the Deep Facial Expression Vector ExtractoR (DeepFEVER), a new deep learning-based approach that learns a visual feature extractor general enough to be applied to any other facial emotion recognition task or dataset. DeepFEVER outperforms state-of-the-art results on the AffectNet and Google Facial Expression Comparison datasets. DeepFEVER’s extracted features also generalise extremely well to other datasets – even those unseen during training – namely, the Real-World Affective Faces (RAF) dataset.

Emotional expressions are the behaviors that communicate our emotional state or attitude to others. They are expressed through verbal and non-verbal communication. Complex human behavior can be understood by studying physical features from multiple modalities; mainly facial, vocal and physical gestures. Recently, spontaneous multi-modal emotion recognition has been extensively studied for human behavior analysis. In this paper, we propose a new deep learning-based approach for audio-visual emotion recognition. Our approach leverages recent advances in deep learning like knowledge distillation and high-performing deep architectures. The deep feature representations of the audio and visual modalities are fused based on a model-level fusion strategy. A recurrent neural network is then used to capture the temporal dynamics. Our proposed approach substantially outperforms state-of-the-art approaches in predicting valence on the RECOLA dataset. Moreover, our proposed visual facial expression feature extraction network outperforms state-of-the-art results on the AffectNet and Google Facial Expression Comparison datasets.