Robust machine learning segmentation for large-scale analysis of heterogeneous clinical brain MRI datasets.

Every year, millions of brain MRI scans are acquired in hospitals, which is a figure considerably larger than the size of any research dataset. Therefore, the ability to analyze such scans could transform neuroimaging research. Yet, their potential remains untapped since no automated algorithm is robust enough to cope with the high variability in clinical acquisitions (MR contrasts, resolutions, orientations, artifacts, and subject populations). Here, we present SynthSeg+, an AI segmentation suite that enables robust analysis of heterogeneous clinical datasets. In addition to whole-brain segmentation, SynthSeg+ also performs cortical parcellation, intracranial volume estimation, and automated detection of faulty segmentations (mainly caused by scans of very low quality). We demonstrate SynthSeg+ in seven experiments, including an aging study on 14,000 scans, where it accurately replicates atrophy patterns observed on data of much higher quality. SynthSeg+ is publicly released as a ready-to-use tool to unlock the potential of quantitative morphometry.

Monitoring and Control Framework for IoT, Implemented for Smart Agriculture.

To mitigate the effects of the lack of IoT standardization, including scalability, reusability, and interoperability, we propose a domain-agnostic monitoring and control framework (MCF) for the design and implementation of Internet of Things (IoT) systems. We created building blocks for the layers of the five-layer IoT architecture and built the MCF's subsystems (monitoring subsystem, control subsystem, and computing subsystem). We demonstrated the utilization of MCF in a real-world use-case in smart agriculture, using off-the-shelf sensors and actuators and an open-source code. As a user guide, we discuss the necessary considerations for each subsystem and evaluate our framework in terms of its scalability, reusability, and interoperability (issues that are often overlooked during development). Aside from the freedom to choose the hardware used to build complete open-source IoT solutions, the MCF use-case was less expensive, as revealed by a cost analysis that compared the cost of implementing the system using the MCF to obtain commercial solutions. Our MCF is shown to cost up to 20 times less than normal solutions, while serving its purpose. We believe that the MCF eliminated the domain restriction found in many IoT frameworks and serves as a first step toward IoT standardization. Our framework was shown to be stable in real-world applications, with the code not incurring a significant increase in power utilization, and could be operated using common rechargeable batteries and a solar panel. In fact, our code consumed so little power that the usual amount of energy was two times higher than what is necessary to keep the batteries full. We also show that the data provided by our framework are reliable through the use of multiple different sensors operating in parallel and sending similar data at a stable rate, without significant differences between the readings. Lastly, the elements of our framework can exchange data in a stable way with very few package losses, being able to read over 1.5 million data points in the course of three months.

Dual domain distribution disruption with semantics preservation: Unsupervised domain adaptation for medical image segmentation.

Recent unsupervised domain adaptation (UDA) methods in medical image segmentation commonly utilize Generative Adversarial Networks (GANs) for domain translation. However, the translated images often exhibit a distribution deviation from the ideal due to the inherent instability of GANs, leading to challenges such as visual inconsistency and incorrect style, consequently causing the segmentation model to fall into the fixed wrong pattern. To address this problem, we propose a novel UDA framework known as Dual Domain Distribution Disruption with Semantics Preservation (DDSP). Departing from the idea of generating images conforming to the target domain distribution in GAN-based UDA methods, we make the model domain-agnostic and focus on anatomical structural information by leveraging semantic information as constraints to guide the model to adapt to images with disrupted distributions in both source and target domains. Furthermore, we introduce the inter-channel similarity feature alignment based on the domain-invariant structural prior information, which facilitates the shared pixel-wise classifier to achieve robust performance on target domain features by aligning the source and target domain features across channels. Without any exaggeration, our method significantly outperforms existing state-of-the-art UDA methods on three public datasets (i.e., the heart dataset, the brain dataset, and the prostate dataset). The code is available at https://github.com/MIXAILAB/DDSPSeg.