Motivation: Infection (bacteria in the wound) and ischemia (insufficient blood supply) in Diabetic Foot Ulcers (DFUs) increase the risk of limb amputation. Goal: To develop an image-based DFU infection and ischemia detection system that uses deep learning. Methods: The DFU dataset was augmented using geometric and color image operations, after which binary infection and ischemia classification was done using the EfficientNet deep learning model and a comprehensive set of baselines. Results: The EfficientNets model achieved 99% accuracy in ischemia classification and 98% in infection classification, outperforming ResNet and Inception (87% accuracy) and Ensemble CNN, the prior state of the art (Classification accuracy of 90% for ischemia 73% for infection). EfficientNets also classified test images in a fraction (10% to 50%) of the time taken by baseline models. Conclusions : This work demonstrates that EfficientNets is a viable deep learning model for infection and ischemia classification.

Smartphone wound image analysis has recently emerged as a viable way to assess healing progress and provide actionable feedback to patients and caregivers between hospital appointments. Segmentation is a key image analysis step, after which attributes of the wound segment (e.g. wound area and tissue composition) can be analyzed. The Associated Hierarchical Random Field (AHRF) formulates the image segmentation problem as a graph optimization problem. Handcrafted features are extracted, which are then classified using machine learning classifiers. More recently deep learning approaches have emerged and demonstrated superior performance for a wide range of image analysis tasks. FCN, U-Net and DeepLabV3 are Convolutional Neural Networks used for semantic segmentation. While in separate experiments each of these methods have shown promising results, no prior work has comprehensively and systematically compared the approaches on the same large wound image dataset, or more generally compared deep learning vs non-deep learning wound image segmentation approaches. In this paper, we compare the segmentation performance of AHRF and CNN approaches (FCN, U-Net, DeepLabV3) using various metrics including segmentation accuracy (dice score), inference time, amount of training data required and performance on diverse wound sizes and tissue types. Improvements possible using various image pre- and post-processing techniques are also explored. As access to adequate medical images/data is a common constraint, we explore the sensitivity of the approaches to the size of the wound dataset. We found that for small datasets (<; 300 images), AHRF is more accurate than U-Net but not as accurate as FCN and DeepLabV3. AHRF is also over $1000\times$ slower. For larger datasets (>300 images), AHRF saturates quickly, and all CNN approaches (FCN, U-Net and DeepLabV3) are significantly more accurate than AHRF.

Mobile and tablet apps, ubiquitous and pervasive computing, wearable computing and domotics for health.

Diabetes mellitus is a serious chronic disease that affects millions of people worldwide. In patients with diabetes, ulcers occur frequently and heal slowly. Grading and staging of diabetic ulcers is the first step of effective treatment and wound depth and granulation tissue amount are two important indicators of wound healing progress. However, wound depths and granulation tissue amount of different severities can visually appear quite similar, making accurate machine learning classification challenging. In this paper, we innovatively adopted the fine-grained classification idea for diabetic wound grading by using a Bilinear CNN (Bi-CNN) architecture to deal with highly similar images of five grades. Wound area extraction, sharpening, resizing and augmentation were used to pre-process images before being input to the Bi-CNN. Innovative modifications of the generic Bi-CNN network architecture are explored to improve its performance. Our research generated a valuable wound dataset. In collaboration with wound experts from University of Massachusetts Medical School, we collected a diabetic wound dataset of 1639 images and annotated them with wound depth and granulation tissue grades as labels for classification. Deep learning experiments were conducted using holdout validation on this diabetic wound dataset. Comparisons with widely used CNN classification architectures demonstrated that our Bi-CNN fine-grained classification approach outperformed prior work for the task of grading diabetic wounds.

The risk of COVID-19 transmission increases when an uninfected person is less than 6 ft from an infected person for longer than 15 minutes. Infectious disease experts working on the COVID-19 pandemic call this high-risk situation being Too Close for Too Long (TCTL). Consequently, the problem of detecting the TCTL situation in order to maintain appropriate social distance has attracted considerable attention recently. One of the most prominent TCTL detection ideas being explored involves utilizing the Bluetooth Low-Energy (BLE) Received Signal Strength Indicator (RSSI) to determine whether the owners of two smartphones are observing the acceptable social distance of 6 ft. However, using RSSI measurements to detect the TCTL situation is extremely challenging due to the significant signal variance caused by multipath fading in indoor radio channel, carrying the smartphone in different pockets or positions, and differences in smartphone manufacturer and type of the device. In this study we utilize the Mitre Range Angle Structured (MRAS) Private Automated Contact Tracing (PACT) dataset to extensively evaluate the effectiveness of Machine Learning (ML) algorithms in comparison to classical estimation theory techniques to solve the TCTL problem. We provide a comparative performance evaluation of proximity classification accuracy and the corresponding confidence levels using classical estimation theory and a variety of ML algorithms. As the classical estimation method utilizes RSSI characteristics models, it is faster to compute, is more explainable, and drives an analytical solution for the precision bounds proximity estimation. The ML algorithms, Support Vector Machines (SVM), Random Forest, and Gradient Boosted Machines (GBM) utilized thirteen spatial, time-domain, frequency-domain, and statistical features extracted from the BLE RSSI data to generate the same results as classical estimation algorithms. We show that ML algorithms can achieve 3.60%~19.98% better precision, getting closer to achievable bounds for estimation.

Breast cancer (BC) is currently the most common form of cancer diagnosed worldwide with an incidence estimated at 2.26 million in 2020. Additionally, BC is the leading cause of cancer death. Many subtypes of breast cancer exist with distinct biological features and which respond differently to various treatment modalities and have different clinical outcomes. To ensure that sufferers receive lifesaving patients-tailored treatment early, it is crucial to accurately distinguish dangerous malignant (ductal carcinoma, lobular carcinoma, mucinous carcinoma, and papillary carcinoma) subtypes of tumors from adenosis, fibroadenoma, phyllodes tumor, and tubular adenoma benign harmless subtypes. An excellent automated method for detecting malignant subtypes of tumors is desirable since doctors do not identify 10% to 30% of breast cancers during regular examinations. While several computerized methods for breast cancer classification have been proposed, deep convolutional neural networks (DCNNs) have demonstrated superior performance. In this work, we proposed an ensemble of four variants of DCNNs combined with the support vector machines classifier to classify breast cancer histopathological images into eight subtypes classes: four benign and four malignant. The proposed method utilizes the power of DCNNs to extract highly predictive multi-scale pooled image feature representation (MPIFR) from four resolutions (40×, 100×, 200×, and 400×) of BC images that are then classified using SVM. Eight pre-trained DCNN architectures (Inceptionv3, InceptionResNetv2, ResNet18, ResNet50, DenseNet201, EfficientNetb0, shuffleNet, and SqueezeNet) were individually trained and an ensemble of the four best-performing models (ResNet50, ResNet18, DenseNet201, and EfficientNetb0) was utilized for feature extraction. One-versus-one SVM classification was then utilized to model an 8-class breast cancer image classifier. Our work is novel because while some prior work has utilized CNNs for 2- and 4-class breast cancer classification, only one other prior work proposed a solution for 8-class BC histopathological image classification. A 6B-Net deep CNN model was utilized, achieving an accuracy of 90% for 8-class BC classification. In rigorous evaluation, the proposed MPIFR method achieved an average accuracy of 97.77%, with 97.48% sensitivity, and 98.45% precision on the BreakHis histopathological BC image dataset, outperforming the prior state-of-the-art for histopathological breast cancer multi-class classification and a comprehensive set of DCNN baseline models.