Recent advances in artificial intelligence (AI) have shown promising results in various image-based systems, improving accuracy and throughput, while reducing latency. All these factors are crucial in healthcare and have generated increased interest in this technology. However, there are also multiple challenges integrating AI in existing systems, such as poor explainability, data imbalance and bias. These challenges affect the reliability of the neural networks used in AI applications. The limitations may significantly affect the quality and cost of medical care by introducing false positive diagnosis. The false positives subsequently lead to increased stress in patients and necessitate additional testing and procedures. Lack of rich data representing all socio-economic groups can also undermine reliable decisions for underrepresented groups. Although various studies discussed techniques that may help with bias mitigation, to the best of our knowledge, no practical experiments have been conducted so far that compare different reweighting approaches using convolutional neural networks (CNN). This work focuses on in-depth explanatory analysis of chest X-ray datasets to understand and quantify the problem of class imbalance and bias. After that, various topologies of binary classifications are compared, followed by practical applications of loss reweighting techniques and comparison of their influence of privileged, underprivileged, and overall population. Experiments proved that high classification accuracy can be achieved using an efficient model topology suitable for embedded devices, making it possible to run locally without the need for cloud processing. Preliminary results showed that performance of the model for the underprivileged class can be improved by 15% if proper weighting factors are obtained and applied during the training procedure.

Estimation of vital signs using image processing techniques have already been proved to have a potential for supporting remote medical diagnostics and replacing traditional measurements that usually require special hardware and electrodes placed on a body. In this paper, we further extend studies on contactless Respiratory Rate (RR) estimation from extremely low resolution thermal imagery by enhancing acquired sequences using Deep Neural Networks (DNN). To perform extensive benchmark evaluation, we acquired two thermal datasets using FLIR^® cameras with a spatial resolution of 80 × 60 and 320 × 240 from 71 volunteers in total. In-depth analysis of the proposed Convolutional-based Super Resolution model showed that for images downscaled with a factor of 2 and then super-resolved using Deep Learning (DL) can lead to better RR estimation accuracy than from original high-resolution sequences. In addition, if an estimator based on a dominating peak in the frequency domain is used, SR can outperform original data for a down-scale factor of 4 and images as small as 20 × 15 pixels. Our study also showed that RR estimation accuracy is better for super-resolved data than for images with color changes magnified using algorithms previously applied in the literature for enhancing vital signs patterns.

As healthcare costs continue to rise, finding affordable and non-invasive ways to monitor vital signs is increasingly important. One of the key metrics for assessing overall health and identifying potential issues early on is respiratory rate (RR). Most of the existing methods require multiple steps that consist of image and signal processing. This might be difficult to deploy on edge devices that often do not have specialized digital signal processors (DSP). Therefore, the goal of this study is to develop a single neural network realizing the entire process of RR estimation in a single forward pass. The proposed solution builds on recent advances in video recognition, capturing both spatial and temporal information in a multi-path network. Both paths process the data at different sampling rates to capture rapid and slow changes that are associated with differences in the temperature of the nostril area during the breathing episodes. The preliminary results show that the introduced end-to-end solution achieves better performance compared to state-of-the-art methods, without requiring additional pre/post-processing steps and signal-processing techniques. In addition, the presented results demonstrate its robustness on low-resolution thermal video sequences that are often used at the embedded edge due to the size and power constraints of such systems. Taking that into account, the proposed approach has the potential for efficient and convenient respiratory rate estimation across various markets in solutions deployed locally, close to end users.

Recent one-shot Neural Architecture Search algorithms rely on training a hardware-agnostic super-network tailored to a specific task and then extracting efficient sub-networks for different hardware platforms. Popular approaches separate the training of super-networks from the search for sub-networks, often employing predictors to alleviate the computational overhead associated with search. Additionally, certain methods also incorporate the quantization policy within the search space. However, while the quantization policy search for convolutional neural networks is well studied, the extension of these methods to transformers and especially foundation models remains under-explored. In this paper, we demonstrate that by using multi-objective search algorithms paired with lightly trained predictors, we can efficiently search for both the sub-network architecture and the corresponding quantization policy and outperform their respective baselines across different performance objectives such as accuracy, model size, and latency. Specifically, we demonstrate that our approach performs well across both uni-modal (ViT and BERT) and multi-modal (BEiT-3) transformer-based architectures as well as convolutional architectures (ResNet). For certain networks, we demonstrate an improvement of up to $4.80x$ and $3.44x$ for latency and model size respectively, without degradation in accuracy compared to the fully quantized INT8 baselines.

One-Shot Neural Architecture Search (NAS) algorithms often rely on training a hardware agnostic super-network for a domain specific task. Optimal sub-networks are then extracted from the trained super-network for different hardware platforms. However, training super-networks from scratch can be extremely time consuming and compute intensive especially for large models that rely on a two-stage training process of pre-training and fine-tuning. State of the art pre-trained models are available for a wide range of tasks, but their large sizes significantly limits their applicability on various hardware platforms. We propose InstaTune, a method that leverages off-the-shelf pre-trained weights for large models and generates a super-network during the fine-tuning stage. InstaTune has multiple benefits. Firstly, since the process happens during fine-tuning, it minimizes the overall time and compute resources required for NAS. Secondly, the sub-networks extracted are optimized for the target task, unlike prior work that optimizes on the pre-training objective. Finally, InstaTune is easy to "plug and play" in existing frameworks. By using multi-objective evolutionary search algorithms along with lightly trained predictors, we find Pareto-optimal sub-networks that outperform their respective baselines across different performance objectives such as accuracy and MACs. Specifically, we demonstrate that our approach performs well across both unimodal (ViT and BERT) and multi-modal (BEiT-3) transformer based architectures.