Abstract:

This PhD thesis focuses on the development of deep learning methods to enhance medical image analysis, with a particular emphasis on improving focus, efficiency and fairness in acute ischemic stroke (AIS). AIS is a cerebrovascular disease that occurs when a cerebral artery becomes occluded, interrupting blood flow to part of the brain. Rapid diagnosis and treatment of AIS are essential to preserving salvageable tissue and minimizing long-term disability. However, not all patients benefit equally from current therapies despite similar clinical and procedural characteristics. Accurately identifying patients likely to benefit remains a major clinical challenge. In this context, neuroimaging, particularly brain Computed Tomography Angiography (CTA), plays a central role in patient triage during the acute stroke phase. The rich anatomical and vascular information present in CTA offers a valuable but complex input for automated image analysis, motivating the use of deep learning to support clinical decision-making.

This thesis explores strategies to fully utilize CTA data using deep learning methods tailored for AIS. The first contribution proposes strategies to guide neural networks toward vascular structures extracted from CTA while retaining contextual parenchymal information. Several methods to combine vascular segmentations and CTA data are evaluated, including an attention-inspired mechanism designed to enhance the model’s focus on clinically relevant features for AIS due to large vessel occlusion (LVO) detection. The second contribution introduces Learnable 3D Pooling (L3P), a novel convolutional neural network-based module that compresses
3D medical images into 2D feature maps, enabling efficient, lightweight, and interpretable models. L3P-based architectures are validated across multiple tasks, including LVO detection in CTA, as well as brain age prediction from 3D T1-weighted MRI. In all cases, L3P maintains competitive performance compared to fully 3D networks, while significantly reducing computational demands and enhancing feature interpretability. The third contribution addresses the issue of hidden confounders in medical imaging pipelines. A controlled experimental framework is proposed to simulate and analyze the effects of confounding variables in clinically relevant classification tasks. Using large ensembles of bootstrapped models, measurable distributional patterns, such as inflated model performance, reduced performance variability and convergence of training and validation metrics, are identified as indicators of confounder-driven learning. This enables the development of a practical, unsupervised method to flag potential hidden biases, even when the nature of the confounder is unknown.

The proposed approaches are broadly applicable across medical imaging modalities and disease domains, with particular relevance to neuroimaging. Designed with generalizability in mind, these methods have demonstrated utility beyond stroke, highlighting their potential for diverse clinical contexts. Collectively, the contributions of this thesis advance the development of deep learning tools with efficiency, focus, and fairness.

The complete doctoral thesis can be consulted in the official PhD repository of the University of Girona (DUGi-doc):  https://dugi-doc.udg.edu

The post Doctoral Thesis: Deep Learning for Stroke Imaging: Enhancing Focus, Efficiency, and Fairness appeared first on Vicorob.

Abstract:

Skin cancer remains one of the most prevalent and deadly forms of cancer worldwide, with melanoma alone accounting for over 330,000 new cases and nearly 60,000 deaths in 2022. Early detection is critical, as survival rates drop dramatically from 99% to just 30% once the cancer has metastasized.

This thesis was conducted within the Computer Vision and Robotics (VICOROB) group and the European Union iToBoS project. It advances AI-driven tools for skin cancer diagnosis with a focus on clinical and dermoscopic image analysis to support a two-tiered screening workflow.

In the iToBoS diagnostic pipeline, first a full-body scan is performed to capture clinical images of all visible skin lesions. Then, dermoscopic images are acquired
for lesions identified as suspicious during the initial clinical assessment, allowing a more detailed examination.

The presented research contributes to the workflow and advances the field through four key contributions. First, we review clinical image-based diagnosis, identifying successful methods and unresolved challenges, then deploy a pre-trained model for clinical image classification. Second, we design compact deep learning models for real-time dermoscopic melanoma detection by integrating attention mechanisms to improve accuracy while reducing computational costs. Third, we expand the dermoscopic diagnosis to include other types of skin cancer and enhance performance through clinically structured labeling. The resulting ensemble model outperforms existing approaches while balancing sensitivity and specificity for realworld use. Finally, we integrate vision-language models to provide dermoscopiclevel explainability, ensuring transparent and interpretable diagnoses.

Together, these contributions enable accurate, scalable, and clinically viable skin cancer detection systems across two imaging modalities with the final objective of improving patient outcomes and reducing mortality rate.

The post Doctoral Thesis: Development of Intelligent Systems for Skin Cancer Diagnosis appeared first on Vicorob.

Abstract:

Locating an Autonomous Underwater Vehicle (AUV) is a complex problem since most of the common sensors applied in ground or aerial robotics experience a significant reduction in their capabilities underwater due to the strong attenuation of electromagnetic waves
within the water. Conventional underwater navigators use acoustic positioning systems that share information through water. However, the deployment of these systems is costly. A more inexpensive option is to fuse the AUV proprioceptive measurements with the measurements of the robot surroundings, solving the Simultaneous Localization and Mapping (SLAM) problem. Although during the last decades a great research effort has been made on this problem,
some fundamental questions concerning its formulation remain open. Aspects such as using Lie groups to model robot poses, applying a factor graph to model the joint probability distribution of the problem, developing an online solver algorithm, or reaching a tightly coupled
estimation have not found a consensus until recent years.

To enhance the surveying capabilities of an AUV, this thesis focuses on the development of a state-of-the-art AUV navigator, transferring recent findings from generalist robotics to the underwater field. Thus, a tightly coupled and graph-based underwater navigator is proposed that properly applies Lie algebra when necessary. The navigator is based on acoustic range sensors that provide point cloud measurements of the robot’s surroundings.

The main contributions of this thesis are two, properly presented in the four articles that form this document. The first one is the development of a scan matching algorithm based on Gaussian Mixture Models (GMMs) to represent the sensor noise projected to the scan that
returns the uncertainty associated with the alignment result, an essential metric in SLAM problems. In addition, the Bayesian-GMM algorithm is first introduced to learn a GMM from a point cloud. The second main contribution of this thesis is the development of an algorithm to jointly preintegrate Inertial Measurement Unit (IMU) and Doppler Velocity Log (DVL) measurements to reach a tightly coupled estimation in an underwater Graph SLAM problem. Moreover, it allows compensating the preintegrated measurement from the Earth rotation
rate measured by high grade IMUs. Both algorithms are formulated considering Lie algebra to properly manipulate robot states and sensor measurements, introducing the SEN(3) group to jointly preintegrate IMU and DVL measurements.

Finally, this thesis also includes field experiments that prove the performance of the two proposed underwater navigators, one applying a Mechanical Profiling Sonar (MPS) and the other using a Mechanical Scanning MultiBeam EchoSounder (MS-MBES). The software architecture developed to implement both navigators is also presented, which provides a graphbased navigation framework to implement other navigators applying other sensor modalities.

The post Doctoral Thesis: Graph-based underwater localization techniques considering a rigorous on Lie group formulation appeared first on Vicorob.

Valerio presented the paper titled “Collision Avoidance with Adaptive Potential Fields for Underwater Vehicles Using Omnidirectional Vision”. The work, co-authored by Aurora Bottino (a Master student from Università degli Studi di Genova, Italy), and members of our research group Eduardo Ochoa, Rafael Garcia and Nuno Gracias, was accepted to the main technical session. It presents an algorithm designed to provide access to remotely operated underwater vehicles (ROVs) to any non-skilled human operator. Moving an underwater vehicle close to underwater structures to capture images is a challenging task that compromises the safety of the vehicle. For this reason, highly skilled human operators are chosen for this task. We developed a collision avoidance system that uses visual information of its surroundings from an omnidirectional camera to correct the robot’s direction of travel to maintain ROV/AUV safety while in proximity to obstacles, such that anybody can operate one without there being any collision risk.

Alaaeddine presented the paper titled “A Docking Station for the Girona I-AUV: Validation by Simulation”. The work, co-authored by Joan Esteba (former PhD at CIRS, now post-doctoral researcher at New York University Abu Dhabi, United Arab Emirates) and members of our research group Patryk Cieslak and Pere Ridao, was accepted to the main technical session. The paper showcases the development process of the docking station for the Girona I-AUV. It evaluated several docking station proposals to choose the most suitable design for the vehicle. The assessment of the proposed designs focused on three critical parameters: structural size, manufacturing cost and operational feasibility. The vertical docking design was ultimately chosen as it demonstrated an optimal balance between these requirements. It also includes several critical innovations: a dual-axis guidance system accommodating ±0.25m positional and ±25◦ yaw tolerances, a secure sliding lock mechanism for underwater station-keeping, and integrated data/power transfer capabilities.

The OCEANS conference was a nice opportunity to share the work we have been doing at Girona with the rest of the underwater robotics and marine science community, to learn what other researchers are doing, and to foster new collaborations with other research institutions.

The post Participation at OCEANS 2025 in Brest appeared first on Vicorob.

Abstract:

This thesis explores how advanced artificial intelligence techniques, specifically deep learning, can improve the analysis of brain scans (MRI) for people with multiple sclerosis (MS) in the clinical practice.

The study focuses on three key areas. First, it introduces a new AI tool designed to accurately and consistently isolate the brain from the surrounding tissues in MRI scans. This is crucial for many analyses and can improve the accuracy of brain volume measurements, which are important for tracking disease progression. Secondly, the research develops a method to generate synthetic brain scans from existing ones. This can help improve the detection of MS lesions (areas of brain damage) while potentially reducing the need for expensive and time-consuming MRI scans.

Finally, the study investigates the practical challenges of bringing these AI tools into real-world clinical use. This includes navigating regulations and ensuring the safety and effectiveness of these technologies for patients.

In summary, this research aims to improve the diagnosis and management of MS by developing and implementing innovative AI solutions for analyzing brain MRI scans.

The post Industrial Doctoral Thesis: Applications of deep learning techniques in Magnetic Resonance Imaging for Multiple Sclerosis: from research innovations to clinical implementation appeared first on Vicorob.

Abstract:

The interest in the use of autonomous underwater vehicles (AUVs) has increased in recent decades. While former research focused on underwater exploration for sea bottom map- ping (bathymetries, sonar, and photo mosaics), it evolved soon into 3D optical reconstruction and offshore infrastructure inspection. Progress in these areas has sparked the interest of the community in employing AUVs for intervention tasks, thereby replacing remotely operated vehicles (ROVs) and manned submersibles with intervention autonomous underwater vehicles (I-AUVs). This substitution offers the potential to automate tasks, improving efficiency and repeatability while reducing costs, time, and logistics. However, autonomous intervention underwater is challenging. It requires the joint control of a heterogeneous multibody sys- tem composed of the AUV and the manipulators, which have significant differences in terms of control and accuracy.

Most intervention tasks, such as object grasping or valve turning, require centimeter accuracy in the position of the end effector. This accuracy is severely affected by a chain of errors, beginning with the navigation error and continuing with the cal- ibration errors of the involved systems, including inaccuracies in the positions of the cameras, lasers, and manipulators, joint calibration errors, and other uncertainties within the system. Another challenge is the manipulation of bulky objects which are difficult, if not impossi- ble, to satisfy with a single vehicle. Most probably, future autonomous intervention systems will be multi-robot. This poses new problems to solve, like the joint control of a team of I-AUVs coordinated through low bandwidth communication channels. Finally, it is necessary to root the autonomous underwater intervention research to the actual needs of field appli- cations. This thesis is a contribution along these lines. It aims to advance the autonomous underwater intervention state of the art to increase the autonomy of I-AUVs for inspection, maintenance, and repair (IMR) tasks in offshore infrastructures. First, a new framework is proposed to calibrate the intrinsic/extrinsic parameters of the I-AUVs components, using ro- bust modeling of the minimization equations leveraging Lie theory. Then, the Task Priority redundancy control algorithm is enhanced to control two I-AUVs, communicating through a low-rate communications channel to transport a bulky object.

Finally, an effort is made to study the actual capabilities of I-AUVs to face field applications in the area of offshore re- newable energies. A Task Priority algorithm supporting admittance control is used to control an I-AUV performing non-destructive inspection for cathodic protection on a floating semi- submersible windmill structure. Throughout the thesis, all the works present both simulation and experimental results, validating the efficiency and potential of the proposed solutions.

https://www.udg.edu/en/ed/tesis-doctorals/llista-de-tesis/codi/350130813

The post Doctoral Thesis: Enhancing Underwater Operations through Advanced Autonomous Manipulation appeared first on Vicorob.

Abstract:

Underwater robotics has undergone significant development in recent years. It has been applied to a wide range of sectors, such as the mapping of areas of interest, the collection of scientific data, or the Inspection Maintainance and Repair (IMR) tasks for the energy sector (oil and gas, renewable energies, etc.). Nowadays, Remotely Operated Vehicles play a leading role in these fields and are gradually being replaced by Autonomous Underwater Vehicles (AUVs).

In the coming years, the market will need AUVs deployed for long term in strategic loca- tions, such as oshore wind farms. To achieve this goal, a key factor is the development of Docking Station (DS) where robots can be stationed, charge their batteries, and have a stable channel for fast communication. With this in mind, this thesis focuses on the development of new technologies for the Long Term Deployment (LTD) of non-holonomic AUVs at sites of interest.

The work began with a review of the state of the art. Next, a new metric for scoring docking success was proposed and used for the comparison of dierent strategies. Then, a new docking algorithm that takes into account the ocean current was proposed, simulated, and compared to methods in the literature; with promising results. At this point, a new funnel-based DS, which can be self-aligned with the ocean current to simplify the docking process, was designed and implemented. Finally, the proposed DS and docking algorithm were validated at sea using Sparus II AUV equipped with an inverse Ultra-Short BaseLine (USBL) system for the DS localization. The results demonstrate the validity of the proposal and pave the way for applications requiring the LTD of AUVs.

https://www.udg.edu/en/ed/tesis-doctorals/llista-de-tesis/codi/350130813

The post Doctoral Thesis: Enhancing the AUV long-term deployment: Non-holonomic AUV autonomous docking using acoustics in a funnel-shaped docking station appeared first on Vicorob.

Supervised by Dr. Xavier Lladó / Dr. Arnau Oliver / Dr. Sergi Valverde

Abstract

The post Doctoral Thesis: Deep learning methods for extraction of neuroimage markers in the prognosis of brain pathologies appeared first on Vicorob.

Supervised by Dr.Pere Ridao Rodríguez / Dr. Josep Forest Collado

Abstract

The post Doctoral Thesis: Underwater 3D Sensing using Structured Light: Development of an Underwater Laser Scanner and a Non-Rigid Point Cloud Registration Method appeared first on Vicorob.

Supervised by Dr.Pere Ridao / Dr.Nuno Gracias

Abstract

In recent decades, the rapid development of intelligent vehicle and 3D scanning technologies has led to a growing interest in applications based on 3D point data processing, with many applications such as augmented reality or robot manipulation and obstacle avoidance, scene understanding, robot navigation, tracking and assistive technology among others, requiring an accurate solution for the 3D pose of the recognized objects. Thus object recognition is becoming an important topic in computer vision, where machine vision and robotics techniques are becoming key players.

In this thesis work, the main objective is to develop a semantic mapping method by integrating a 3D object recognition pipeline with a feature-based SLAM system, in order to assist autonomous underwater interventions in the near future.

To this end, the work proposed in this paper targets three axes. First, it aims to compare the performance of 3D global descriptors within the state of the art, focusing on those based on point clouds and targeted at real-time object recognition applications. For this purpose, we selected a set of test objects representative of Inspection, Maintenance and Repair (IMR) applications and whose shape is usually known a priori. Their CAD models were used to: 1) create a data base of synthetic object views used as a priori knowledge, and 2) simulate the point clouds that would be gathered during the scanning under realistic conditions, with added noise and varying resolution. Extensive experiments were performed with both virtual scans and real data collected with an AUV equipped with a fast laser scanner developed at our research centre.

The second goal of our work was to use a real-time laser scanner mounted on an AUV to detect, identify, and locate objects in the robot’s environment, with the aim of allowing an intervention Autonomous Underwater Vehicle (I-AUV) to know what manipulation actions could be performed on each object. This goal was tackled by the design and development of a 3D object recognition method for uncolored point clouds (laser scans) using point features. The algorithm uses a database of partial views of the objects stored as point clouds. The recognition pipeline includes 5 stages: 1) Plane segmentation, 2) Pipe detection, 3) Semantic Object-segmentation, 4) Feature-based Object Recognition and 5) Bayesian estimation. To apply Bayesian estimation, it is necessary to track objects across scans. For this purpose, the Inter-distance Joint Compatibility Branch and Bound (IJCBB) data association algorithm was proposed based on the distances between objects. The performance of the method was tested using a dataset of the inspection of a pipe infrastructure made of PVC objects connected by pipes. The structure is representative of those commonly used by the offshore industry. Experimental results show that Bayesian estimation improves the recognition performance with respect to the case where only the 1 descriptor is used. The inclusion of semantic information about object pipe connectivity further improves recognition performance.

The final goal of the thesis, consists of integrating the 3D object recognition system with a feature-based SLAM system to implement a semantic map providing the robot with information about the location and the type of objects in its surroundings. The SLAM improved both the accuracy and reliability of pose estimates of the robot and the objects. This is especially important in challenging scenarios where significant changes in viewpoint and appearance arise.

The post DOCTORAL THESIS: Automated 3D object recognition in underwater scenarios for manipulation appeared first on Vicorob.