Rehabilitation Engineering and Robotics
Assessing the correctness of physiotherapeutical tasks using IMU data only
Supervisor: Prof. Simona Ferrante
Collaborations: Prof. Thomas Seel and Timo Kuhlgatz, Institute of Mechatronic Systems, Leibniz University Hanover
Aim: Physiotherapy is most promising when the exercises are carried out regularly. However, due to the acute shortage of staff in the healthcare sector, supervised physiotherapy appointments are rare. Performing the exercises independently is therefore of great importance for the success of the therapy. To provide automatic feedback that supports patients in correctly performing the exercise, the patient’s motions can be measured using wearable inertial sensors and assessed in real time using suitable classification algorithms
Requirements: Python programming and interest in physiotherapeutic tasks
DisturbanceRejection due to Non-Rigid IMU Attachment
Supervisor: Prof. Simona Ferrante
Collaborations: Prof. Thomas Seel, Institute of Mechatronic Systems, Leibniz University Hanover; Simon Bachhuber, Friedrich-Alexander-Universität Erlangen-Nürnberg
Aim: Soft-tissue motion (STM) describes the motion of skin relative to the bone structure. Inertial Motion Tracking (IMT) is the field dedicated to estimating dynamic motion of humans (or robots, vehicles etc.) in 3D space. In humans, dynamic movements lead to STM, which then degrades the IMT results, since the sensors are attached to the skin. Can we use machine learning methods to implicilty learn kinematic constraints and successfully eliminate the impact of SMT and recover the underlying rigid body motion?
Requirements: Python programming and Familarity with deep learning frameworks and working on HPC
Co-design of hybrid robotic solutions for the upper limb rehabilitation of neurological patients @Aalborg University (DK)
Supervisor: Prof. Simona Ferrante
Collaborations: Erika G Spaich Department of Health Science and Technology, Aalborg University
Aim: In the recent year the impact of robotics in the rehabilitation of neurological patients is becoming evident. This thesis is aimed at integrating functional electrical stimulation in the control loop of the armeo spring passive exoskeleton in order to further enhance the benefits of rehabilitation combining the two methodologies. The thesis will include: 1) the integration of a stimulation device in the setup, 2) the definition with clinicians of training exercise to be implemented, 3) the development of the FES hybrid control system to support the exercises; the development of a GUI for the clinician and the patient; 4) the testing and evaluation of the results (feasibility, usability…)
Requirements: Interest in learning Matlab and Python
Robust simultaneous recognition of appearance and type of movements
Supervisor: Prof. Simona Ferrante
Collaborations: Prof. Thomas Seel, Institute of Mechatronic Systems, Leibniz University Hannover; Ive Weygers, Friedrich-Alexander-Universität Erlangen-Nürnberg
Aim: Accurate classification of movements such as daily living activities and functional movements is a crucial requirement for a growing number of digital health applications. Classification of isolated recorded movements has been well studied and achieved in many application domains. However, this task becomes challenging with long continuous motion streams of data where the duration and type of the task may vary. The task now requires simultaneous recognition of the occurrence & classification of the specific task, in a robust and ideally near real-time manner using mobile phone data.
Requirements: Python programming and Android SDK for mobile and smartphone applications
Virtual reality for neurological patients rehabilitation
Supervisor: Prof. Simona Ferrante
Co-Supervisor: Linda Greta Dui
Collaborations: IRCCS Istituti Clinici Scientifici Maugeri, Milano
Aim: Patients who suffer from a neuroogical disease show difficulties in performing movements, e.g., «freezing of gate» in Parkinson Disease. It is known that specific cues – e.g., a rhythmical pacing – could help in starting the movement and keeping it fluid. However, it is still unknown the role of psychological distress. All these elements can be combined and tested in a virtual reality scene, that allows to manipulate variables in an immersive virtual world, whilst keeping the patients monitored.
The aim of the thesis work is (1) to develop an app for a VR experience to be paired with movement analysis technologies, and (2) to test it in order to assess usability and the effect of different modifiers on movement performance.
Requirements: Interest in learning C# and Unity
Multimodal analysis of the supraspinal locomotor control in patients with Parkinson
Supervisor: Prof. Alessandra Pedrocchi, Prof. Ioannis U. Isaias
Co-Supervisor: Ing. Chiara Palmisano Ph.D, Ing. Laura Caffi
Collaborations: Julius-Maximilians-Universität (Würzburg, Germany), Centro Parkinson e Parkinsonismi ASST G. Pini-CTO (Milano, Italy)
Aim: Gait modulation is a complex motor task involving multiple cortical and subcortical brain areas. In Parkinson’s disease (PD), gait disturbances result in a dramatic reduction in quality and quantity of life. The candidate will analyze FDG-PET data and EEG recordings of PD patients collected during resting and walking in a fully immersive virtual reality environment specifically designed to induce and study gait modulation. Project phases: Literature review on PD, supraspinal locomotor network, PET and EEG data analysis methods, Data recordings in a small subset of patients, Data analysis and statistics, Presentation of the data at a national conference (if the candidate proves merit), Thesis and article writing
Requirements: Basic knowledge of MATLAB ,Proactive, independent and creative attitude, Availability to spend about 3-4 months in Würzburg (housing and travel costs reimbursed), Priority will be given to students interested in a future Ph.D.
Long-term home monitoring of subcortical neural signals and motor behavior in Parkinson’s patients
Supervisor: Prof. Alessandra Pedrocchi, Prof. Ioannis U. Isaias
Co-Supervisor: Ing. Chiara Palmisano Ph.D, Ing. Laura Caffi
Collaborations: Julius-Maximilians-Universität (Würzburg, Germany), Centro Parkinson e Parkinsonismi ASST G. Pini-CTO (Milano, Italy)
Aim: The student will analyze neural data recorded from the subthalamic nucleus over months in patients with Parkinson’s disease (PD) and deep brain stimulation (DBS). The goal is to correlate the neural signals with disease symptoms, response to neurostimulation therapy, and main activities of daily living (e.g., walking, talking, sleeping, etc.) for the development of novel stimulation paradigms (adaptive DBS).
Project phases: Literature review on PD, DBS, and home monitoring techniques, Training on wearable sensors and DBS recordings, Data recordings in a small subset of patients, Data analysis and statistics, Presentation of the data at a national conference (if the candidate proves merit), Thesis and article writing.
Requirements: Basic knowledge of MATLAB ,Proactive, independent and creative attitude, Availability to spend about 3-4 months in Würzburg (housing and travel costs reimbursed), Priority will be given to students interested in a future Ph.D.
Study of the impact of robotic therapy in children with autism 
Supervisor: Prof. Emilia Ambrosini, prof Alessandra Pedrocchi
Co-Supervisor: Laura Santos, Gabriele Fassina
Collaborations: Fondazione Don Gnocchi Instituto Superior Técnico – Universidade de Lisboa
Aim: Autism still does not have any cure. Several therapies have been developed using robots but their impact is still not clear. To clarify this aspect, a Randomised Controlled Trial has been conceived and has just started in Fondazione Don Gnocchi. The aim of this work is to analyse the results of the ongoing therapy to draw solid conclusion about its effectiveness.
Project phases:
1. Literature analysis about the Robot Assisted Therapies for autism and their assessment
2. Acquisitions at Fondazione don Gnocchi
3. Analysis of the therapy outcomes regarding motion and cognitive parameters extracted from Azure Kinect.
Requirements: Motivation to interact with the therapists, Good Programming skills in Matlab; availability to move to Fondazione don Gnocchi.
Development of a quantitative tool for the assessment of Autism Spectrum Disorder
Supervisor: Prof. Emilia Ambrosini; prof. Alessandra Pedrocchi
Co-Supervisor: Gabriele Fassina, Laura Santos
Collaborations: Fondazione Don Gnocchi Instituto Superior Técnico – Universidade de Lisboa
Aim: Autism Spectrum Disorder (ASD) is currently assessed through clinical scales and questionnaires which suffer from high inter/intra-subject variability. Hence, research seeks to identify new quantitative and objective assessment tool which could benefit from the new advances in technology (e.g. AI). In this view, Joint Attention, i.e. the capability of sharing attention with a partner, could be analysed quantitatively. This project aims at developing an algorithm to quantify Joint Attention from the videos recorded during the administration of a clinical scale. Project Phases:
1. Literature review on ASD symptoms and diagnosis, Computer Vision tools for body and face recognition;
2. Development of a Deep Learning algorithm to quantify Joint Attention;
3. Acquisitions of sessions involving both Typically Developing children and ASD children to apply the proposed algorithm.
Requirements: Motivation to interact with the therapists, Good Programming skills: Python and Matlab; basic knowledge of Deep Learning.
Design and test of control solutions for upper limb exoskeleton 
Supervisor: Prof. Alessandra Pedrocchi
Co-Supervisor: Beatrice Luciani
Collaborations: Prof. Marta Gandolla – WE-COBOT
Aim: Motorized upper limb exoskeletons have been proposed to support motor rehabilitation and/or to assist disabled people during daily life activities. We have developed and tested a 4DOF upper limb exoskeleton – AGREE. The thesis is aimed at developing and testing control solutions for the support of rehabilitation sesions (i.e., tunneling, force fields, assisted-as-needed solutions, synergy-based control, etc.)
Requirements: Basic knowledge of Robotics,Good knowledge of C/C++ and MATLAB,Interest in learning ROS/MoveIt
Service robotics in nursing home: applications for elderly care
Supervisor: Prof. Alessandra Pedrocchi
Co-Supervisor: Luca Pozzi
Collaborations: Prof. Marta Gandolla – WE-COBOT
Aim: The rapidly growing demand for long-term care will become a hard challenge for the healthcare system. An investigation of the unmet-needs of nursing homes and the possible use-cases for a service robot, highlighted the following main areas of focus: (i) communication and socialization; (ii) cognitive stimulation and entertainment; (iii) physical activity and rehabilitation. In this context, the student activity will consist in the design, implementation and testing of service robot software applications (or new features for existing applications), addressing one or more of the identified area of focus. The activity will be based at the inter-dipartimental WE-COBOT (Wearable and Collaborative Robotics) lab in Lecco
Requirements: Basic knowledge of Robotics,Good knowledge of C/C++ and MATLAB,Interest in learning ROS/MoveIt
Evaluation of a exoskeleton-mediated therapy: RCT study
Supervisor: Prof. Alessandra Pedrocchi
Co-Supervisor: Beatrice Luciani
Collaborations: Prof. Marta Gandolla – WE-COBOT, Casa di Cura del Policlinico «Dezza»
Aim: We have conducted a RCT study with the AGREE exoskeleton – an upper limb motorized 4DOF exoskeleton with neurological patients at Casa di Cura del Policlinico Dezza. We have collected instrumental and clinical data and outcome measures, which need to be processed to evaluate the effectiveness of the inclusion of the robotic-based therapy with respect to the standard care alone.
Requirements: Basic knowledge of MATLAB
Benchmarking of upper limb functions in neurological disorders
Supervisor: Prof. Alessandra Pedrocchi
Co-Supervisor: –
Collaborations: Prof. Marta Gandolla – WE-COBOT, Hospial Los Madronos (Madrid)
Aim: We have developed a benchmarking scheme to evaluate upper limb functions in neurological disorders, and we have tested it on a group of healthy subjects. The scheme foresees a protocol and a list of relevant outcome measures. So that the scheme is effectively useful in the clinical context, this project wants to develop an application that guides the acquisition, data analysis and interpretation.
Requirements: Basic knowledge of MATLAB/C++
Integrating Functional Electrical Stimulation in an upper limb exosuit: a hybrid approach
Supervisor: Prof. Emilia Ambrosini
Co-Supervisor: Elena Bardi, Davide Savona
Collaborations: DMEC
Aim: In the context of rehabilitation and assistance, soft exoskeletons (exosuits) represent a promising technology. Nevertheless, the rehabilitative potential of such devices could be boosted with the integration of FES (Functional Electrical Stimulation). The aim of the thesis is to integrate FES in an exosuit for the upper limb and to develop a hybrid control strategy.
Requirements: Good knowledge in programming (Python, C++)
Integrating voluntary effort in a Hybrid Robotic-FES Control System
Supervisor: Prof. Emilia Ambrosini
Co-Supervisor: Tommaso Del Grossi
Collaborations: Prof. Marta Gandolla – DMEC
Aim: Hybrid robotic rehabilitation systems offer advanced motor recovery approaches for patients with neurological impairments. This thesis aims to develop a control system that integrates the patient’s voluntary effort into the existing framework, composed of impedance control for the exoskeleton and ILC for FES, addressing the challenge of dynamic disturbances introduced by voluntary activity. The system must recalibrate the exoskeleton and FES contributions in real-time to optimize assistance and maximize therapeutic outcomes. By detecting and adapting to voluntary efforts, this control system seeks to enhance patient engagement and improve functional recovery.
Requirements: Basic knowledge of Matlab and Matlab Signal Processing Toolbox, Basic knowledge of C++
Methods for MR Image Segmentation and Quantitative Multiparametric MRI Analysis of Skeletal Muscle
Supervisor: Prof. Ambrosini
Co-Supervisor: Dott. Alfonso Mastropietro
Collaborations: Istituto di Sistemi e Tecnologie Industriali Intelligenti per il Manifatturiero Avanzato (STIIMA) del Consiglio Nazionale delle Ricerche – Milano
Aim: The primary objective of this thesis is to develop and evaluate innovative methods for Magnetic Resonance (MR) image segmentation and quantitative multiparametric-MRI analysis of skeletal muscle. We aim to enhance the accuracy and reliability of identifying and analyzing skeletal muscle tissues, leading to improved diagnostic and treatment strategies for related medical conditions.
Project Phases: 1. Literature Review: Comprehensive study of existing methods and technologies in MR image segmentation and multiparametric MRI quantitative analysis; 2. Methodology Development and Implementation: Creation of advanced algorithms or enhancement of existing ones for more precise skeletal muscle identification and quantitative analysis; 3. Evaluation: Comparative analysis with existing methods using predefined metrics to evaluate performance improvements; 4. Final Analysis & Reporting: Comprehensive assessment of all phases with detailed reporting on findings, improvements, challenges, and recommendations for future research.
Requirements: Knowledge of Python, MATLAB, 3D-Slicer, ITK-SNAP
ParkAGILE: Valutazione cinematica ed elettromiografica di protocolli fisioterapici di movimento per il benessere muscolare dei lavoratori digitali
Supervisor: Prof. Pedrocchi
Co-Supervisor:
Collaborations: https://www.parcfor.com/
Aim: ParcAGILE è un attrezzo fisioterapico, compatto e completo, in grado di prevenire e riabilitare il disturbo muscolo scheletrico di origine professionale.
Un sistema unico e innovativo che consente di eseguire, nelle pause fisiologiche di lavoro, movimenti defaticanti, di allungamento muscolare e di mobilità delle articolazioni principalmente coinvolte dalle attività lavorative. L’obbitettivo della tesi è la valutazione cinematica ed elettromiografica necessaria ad ottenere dei feedback di efficacia sull’uso del dispositivo ParcAGILE. Ad integrazione, si ipotizza la realizzazione di un’applicazione che consenta all’utente di svolgere in autonomia una serie di programmi fisioterapici utilizzando ParcAGILE
Artificial Intelligence for Oncology
Explainability approaches for AI-based immunotherapy predictive models for NSCLC using CT scans
Supervisor: Prof. Alessandra Pedrocchi
Co-Supervisor: Margherita Favali (margherita.favali@polimi.it), Vanja Miskovic, Arsela Prelaj
Collaborations: Istituto Nazionale dei Tumori di Milano (INT)
Aim: Immunotherapy has transformed the treatment of non-small cell lung cancer (NSCLC). However, identifying the best candidates for immunotherapy still remains a challenge. In this context, AI-based models has shown promising results in deriving novel data-driven biomarkers from medical imaging, offering a promising approach to enhance patient selection and treatment stratification. However, despite promising results, the integration of these AI-driven tools into clinical practice has been limited. One of the primary barriers is the lack of interpretability of these models, which undermines clinician trust. The aim of this thesis is to evaluate existing methods for explaining AI-based models and to design novel strategies to enhance the interpretability of these models within the context of radiological images (CT scans).
Project phases: 1. Literature review on current explainability methodologies for radiological images; 2. Implementation of explainability approaches (e.g., saliency maps, Grad-CAM, LIME, SHAP) on AI models trained with CT scan data; 3. Design innovative strategies to better interpret AI decisions on CT scans
Requirements: Good knowledge of Python; Good knowledge of Machine and/or Deep Learning
Explainability approaches for LLM-based named entity recognition of immunotherapy clinical features using electronic health records
Supervisor: Prof. Alessandra Pedrocchi
Co-Supervisor: Federica Corso (federica.corso@istitutotumori.mi.it), Vanja Miskovic, Laura Mazzeo, Arsela Prelaj
Collaborations: Istituto Nazionale dei Tumori di Milano (INT)
Aim: Data retrieval in clinical research is often time-consuming and error-prone. Large Language Models (LLMs) could aid the data retrieval stage by exploiting the information reported in Electronic Health Records (EHRs). This study is conducted within the APOLLO 11 trial, a multicentric Italian study focused on the analysis of advanced non-small cell lung cancer (NSCLC) data. Moreover, model explainability have become significantly more important LLMs are used in critical domains such as healthcare. Captum is an open-source model explainability library providing a variety of interpretability methods proposed in recent literature. In the context of information retrieval, Captum attribution methods could be applied to analyze the input prompt’s effect on the output of LLMs. The aim of this thesis is to evaluate existing methods for explaining LLM-based models and to design novel strategies to enhance the interpretability of LLMs within the context of EHRs.
Project phases: 1. Literature review on current explainability methodologies for Large Language Models; 2. Implementation of explainability approaches (e.g., saliency maps, LIME, SHAP) on LLM models; 3. Design innovative strategies to better interpret LLM decisions on EHRs
Requirements: Good knowledge of Python; Good knowledge of Machine and/or Deep Learning; Basics of Natural Language Processing
Explainable AI for Pan‑Treatment Systemic Immune Profiling in Advanced Non‑Small Cell Lung Cancer
Supervisor: Prof. Alessandra Pedrocchi
Co-Supervisor: Cecilia Silvestri (cecilia.silvestri@istitutotumori.mi.it), Vanja Miskovic, Arsela Prelaj
Collaborations: Istituto Nazionale dei Tumori di Milano (INT)
Aim: Immunotherapy and targeted therapies have revolutionized the treatment of non-small cell lung cancer (NSCLC), yet only a fraction of patients experience long-term benefit, and predicting who they are remains a major clinical challenge. While some immune system features are known to influence response to immunotherapy, much less is known about how these insights apply across different treatment types. This project aims to analyse FACS data from patient blood samples, using AI techniques to uncover immune profiles associated with durable response regardless of the type of therapy. This could help stratify patients, distinguishing those likely to respond well to single modality treatment from those who may require more intensive treatment strategies.
Project Phases: 1. Literature review on current AI methodologies for FACS data; 2. Data preprocessing and data curation; 3 Exploratory analysis; 4. Machine learning modeling; 5. Results interpretation
Requirements: Good knowledge of Python; Good knowledge of Machine and/or Deep Learning
Longitudinal monitoring of cough for lung cancer patients
Supervisor: Prof. Emilia Ambrosini
Co-Supervisor: Chiara Giangregorio
Collaborations: Istituto Nazionale dei Tumori Milano (INT), Arsela Prelaj MD
Aim: Non-small cell lung cancer (NSCLC) is a leading cause of cancer-related mortality, with cough being a key symptom reflecting disease progression and treatment response. This thesis analyzes longitudinal cough data collected via an existing mobile app to assess treatment efficacy and patient quality of life. Cough recordings will be processed using advanced machine learning and deep-learning techniques to extract meaningful patterns and trends. These insights will be correlated with clinical outcomes to provide an objective, non-invasive method for monitoring disease status and treatment toxicity.
Requirements: Knowledge of Python; Knowledge of machine learning (deep learning preferable); Willingness to work in a collaborative environment
Speech emotion recognition for lung cancer patients: an AI-based approach
Supervisor: Prof. Emilia Ambrosini
Co-Supervisor: Chiara Giangregorio, William Bennardo
Collaborations: Istituto Nazionale dei Tumori Milano (INT), Istituto Europeo Oncologico
Aim: Non-small cell lung cancer (NSCLC) is a leading cause of cancer-related mortality worldwide, making the monitoring of disease progression crucial for optimizing treatment strategies and improving patient outcomes. Within the i3Lung project, a mobile app was developed to ecologically monitor the quality of life and treatment progression in lung cancer patients. Emotional comorbidities like anxiety and depression can significantly impact treatment adherence, recovery, and overall outcomes; thus, monitoring emotional states is essential for providing personalized care. To this end, the aim of this thesis is training a machine learning or deep learning model on existing speech data to create an automatic and reliable detector of emotions from free speech. This will allow real-time, non-invasive monitoring of patients’ emotional well-being, facilitating early interventions and improving both psychological and physical health outcomes during treatment.
Requirements: Knowledge of Python; Knowledge of machine learning (deep learning preferable); Willingness to work in a collaborative environment
Digital Health
Machine-Learning-based Solution for Inertial Motion Tracking with Non-Rigid IMU Attachment
Supervisor: Simona Ferrante (simona.ferrante@polimi.it)
Co-Supervisor: Prof. Dr.–Ing. Thomas Seel and Timo Kuhlgatz, Institute of Mechatronic Systems, Leibniz University Hanover, timo.kuhlgatz@imes.uni-hannover.de
Aim: Inertial Motion Tracking (IMT) is the field dedicated to estimating dynamic motion of humans (or robots) in 3D space using inertial sensors. Soft-tissue motion describes the motion of wearable sensors on the skin relative to the underlying bone structure, which degrades the accuracy of IMT estimates. We aim to further develop a novel machine-learning-based IMT approach, which has led to impressive results on a large number of IMT problems with rigid IMU attachment, to mitigate the effect of non-rigid IMU attachment and obtain high accuracy despite soft-tissue motion artifacts. This is achieved by domain randomization and sim-to-real transfer techniques. Ground truth data is obtained from marker-based optical motion capture.
Project phases:
•Literature review
•Familiarization with the machine-learning-based IMT method
(https://github.com/simon-bachhuber/ring_supplementary_material)
•Extension of training data and deep learning method
•Evaluation of the results
Requirements: Good knowledge of Python; Familiarity with deep learning and using HPC clusters.
Online Assessment of Physiotherapy Movements by Inertial Sensor Fusion and Machine Learning Algorithms
Supervisor: Simona Ferrante (simona.ferrante@polimi.it)
Co-Supervisor: Prof. Dr.–Ing. Thomas Seel and Timo Kuhlgatz, Institute of Mechatronic Systems, Leibniz University Hanover, timo.kuhlgatz@imes.uni-hannover.de
Collaborations: GAIA Digital Therapeutics AG, Hamburg
Aim: Physiotherapy is most promising when the exercises are carried out regularly and correctly, while performing the exercises independently and without supervision is of great importance for the success of the therapy. We aim to use wearable inertial sensors to track the posture and motion of the body segments during physiotherapy exercises and develop algorithms that provide online feedback on the correctness or effectiveness of the performed motion. This is achieved by implementing inertial sensor fusion and machine learning algorithms in Python. Ground truth data can be derived using video-based analysis or a marker-based optical motion capture system.
Project phases:
•Literature review
•Develop algorithms for the online time series analysis
•Use these algorithms for motion assessement
•Validate the algorithms
Requirements: Good knowledge of Python; interest in digital physiotherapy solutions.
Prendiciting Immuno-therapy response using AI in NSCLC patients integrating multiomics data
Supervisor: Prof. Alessandra Pedrocchi
Co-Supervisor: Dr. Arsela Prelaj
Collaborations:
Aim: Predicting Immunotherapy (IO) efficacy oin NSCLC is a crucial unmet clinical need. Beyond Programmed Death Ligand-1 (PD-L1) no other biomarkers are approved. However, PD-L1 is quite Imperfect. Thus, AI which is able to integrate high dimensional real world and multiomics data could be a unique approach able To find an algorithm able to predict IO in NSCLC patients. The later will be this thesis aim.
Requirements: Knowledge of Matlab; Knowledge of Python (scikit-learn, TensorFlow/Keras);Basic knowledge of Git and Linux is welcome
Machine Learning to longitudinally monitor graphical abilities, towards the early diagnosis of Dysgraphia
Supervisor: Prof Simona Ferrante
Co-Supervisor: Linda Greta Dui
Collaborations: Indipote(dn)s project, Prof. Cristiano Termine (Università dell’Insubria)
Aim: Disentangling transient handwriting difficulties from Dysgraphia is not a trivial task. To facilitate the process, an observational and empowerment study longitudinally followed 200 children for three years, starting from kindergarten and reaching second grade. The aim of this work is to leverage Machine Learning techniques on longitudinal data to (1) predict the level of risk and (2) evaluate the effectiveness of interventions, towards an early screening of Dysgraphia.
Requirements: Matlab, R or Python knowledge; interest in machine learning
A novel acquisition system for the identification of vocal biomarkers in the differential diagnosis of Parkinson’s Disease, Multiple System Atrophy and Progressive Supranuclear Palsy
Supervisor: Prof Simona Ferrante
Co-Supervisor: Chiara Giangregorio, Alice Taborelli
Collaborations: Istituto Neurologico “Carlo Besta” | Fondazione IRCCS, Milano
Aim: Parkinson’s Disease (PD) is a neurodegenerative disorder often characterized by vocal alterations in addition to motor symptoms. These vocal symptoms include monotone speech, limited variations in pitch and volume, imprecise articulation, speech disfluencies, inappropriate pauses, and a harsh voice quality. Similar vocal symptoms may also occur in atypical parkinsonian disorders (APDs), such as Multiple System Atrophy (MSA) and Progressive Supranuclear Palsy (PSP). The overlap in vocal profiles between these conditions complicates the differential diagnosis, particularly in the early stages, with significant implications for prognosis and treatment. Consequently, there is growing interest in identifying vocal biomarkers that can effectively distinguish APDs from PD at an early stage.
Requirements: Basic knowledge of Matlab; interest in machine learning
Project phases:
- Development of acquisition system
- Acquisition of voice recording from PD and APD patients
- Signal processing and Machine Learning analysis
Machine Learning for premature babies care
Supervisor: Prof Simona Ferrante
Co-Supervisor: Linda Greta Dui
Collaborations: Marco Frontini (Link Up s.r.l.); Valentina Bozzetti (Ospedale San Gerardo, Monza)
Aim: Premature babiesneed special care programs to overcome their frailty condition. Their organs are immature and the healthcare support to assure their survival is often invasive (e.g., ventilation, parenteral nutrition). It is important to investigate the effect of different care strategies to personalize the care to maximize the outcome
Requirements: Interest in machine learning
Development of a quantitative tool for the assessment of Autism Spectrum Disorder
Supervisor: Prof. Emilia Ambrosini; prof. Alessandra Pedrocchi
Co-Supervisor: Gabriele Fassina, Laura Santos
Collaborations: Fondazione Don Gnocchi Instituto Superior Técnico – Universidade de Lisboa
Aim: Autism Spectrum Disorder (ASD) is currently assessed through clinical scales and questionnaires which suffer from high inter/intra-subject variability. Hence, research seeks to identify new quantitative and objective assessment tool which could benefit from the new advances in technology (e.g. AI). In this view, Joint Attention, i.e. the capability of sharing attention with a partner, could be analysed quantitatively. This project aims at developing an algorithm to quantify Joint Attention from the videos recorded during the administration of a clinical scale. Project Phases:
1. Literature review on ASD symptoms and diagnosis, Computer Vision tools for body and face recognition;
2. Development of a Deep Learning algorithm to quantify Joint Attention;
3. Acquisitions of sessions involving both Typically Developing children and ASD children to apply the proposed algorithm.
Requirements: Motivation to interact with the therapists, Good Programming skills: Python and Matlab; basic knowledge of Deep Learning.
Handwriting Analysis in Primary School Children with an IoT Smart Ink Pen
Supervisor: Prof. Simona Ferrante
Co-Supervisor: Simone Toffoli, Linda Greta Dui
Collaborations: Università degli Studi dell’Insubria
Aim: 10 to 30% of pupils in primary school show difficulties in mastering handwriting movement skills, a condition known as dysgraphia. Such difficulties have a negative impact on many aspects, like the child’ self-esteem, personal relationship, academic and employment achievements. The diagnosis is performed during grade 3 and is based on the qualitative observation of the handwritten product. National guidelines suggest the development of quantitative approaches for the analysis of children’s handwriting performance (process analysis) in order to: i) create normative data for Italian children of primary school; ii) compare normally developing pupils and children classified as dysgraphic by current clinical protocols. The thesis aim is the collection of handwriting data from pupils with a smart ink pen developed at Nearlab, followed by data analysis.
Requirements: Knowledge of Matlab and Python; Interest in data acquisition in schools
Automation of a serious game for the early screening of dysgraphia
Supervisor: Prof. Simona Ferrante
Co-Supervisor: Linda Greta Dui
Collaborations: Università degli Studi dell’Insubria
Aim: Disentangling transient handwriting difficulties from Dysgraphia is not a trivial task. To facilitate the process, a serious game has been devised, but data analysis needs to be refined to allow the game to be used in a pre-clinical scenario. The aim of the thesis is to optimise data analysis algorithms to automate the prediction of the risk of dygraphia.
Requirements: Knowledge of Matlab, R or Python; Interest in data analysis and machine learning
Sensor-based protocol and acquisition system for the identification of motor biomarkers in the differential diagnosis of Parkinson’s Disease, Multiple System Atrophy and Progressive Supranuclear Palsy
Supervisor: Prof. Simona Ferrante
Co-Supervisor: Alice Taborelli
Collaborations: Istituto Neurologico “Carlo Besta” | Fondazione IRCCS, Milano
Aim: Parkinson’s Disease (PD) is a neurodegenerative disorder primarily characterized by motor symptoms such as resting tremor, limb rigidity, bradykinesia, and, later, postural instability. To improve early differential diagnosis between PD and atypical parkinsonian disorders (APDs) like Multiple System Atrophy (MSA) and Progressive Supranuclear Palsy (PSP), wearable technologies are being employed. Inertial Measurement Units (IMUs) are used for gait analysis, pull-tests, and stand-sit tests, while sensorized pillows with force sensors monitor patients’ seating posture. These technologies aim to identify motor biomarkers that can differentiate APDs from PD, ultimately enabling early intervention. The aim of this thesis will be to validate the acquisition system, collect data from patients with PD and APDs, and analyze the signals using advanced signal processing techniques.
Requirements: Basic knowledge of Matlab; interest in statistics and data analysis
IoT smart ink pen to evaluate the effect of Focused UltraSound (FUS) treatment in patients with essential tremor
Supervisor: Prof. Simona Ferrante
Co-Supervisor: Simone Toffoli
Collaborations: Istituto Neurologico “Carlo Besta” | Fondazione IRCCS, Milano
Aim: Essential tremor (ET) is the most common movement disorder, affecting an estimated 3% of the population. Focused ultrasound is a noninvasive therapeutic technology that focuses multiple beams of ultrasound energy precisely and accurately on targets deep in the brain without damaging surrounding normal tissue. Pre- and post- treatment assessment are conducted on a cohort of patients recruited at the Besta Institute. Evaluation are conducted through clinical scales and through an instrumented smart ink pen able to extract relevant clinical parameters.
Requirements: Basic knowledge of Matlab; interest in statistics and data analysis
IoT smart inkpen for early detection and monitoring of patients with Parkinson’s disease
Supervisor: Prof. Simona Ferrante
Co-Supervisor: Simone Toffoli, Monica Parati
Collaborations: IRCCS Istituti Clinici Scientifici Maugeri, Milano
Aim: The diagnosis of clinically probable Parkinson’s Disease (PD) in the early stages relies primarily on clinical assessment by a neurologist. When motor symptoms affect the dominant hand, patients may report worsening of handwriting as of the initial symptoms. The aim of this work is to use an IoT smart ink pen for the characterization of PD patients handwriting, with the final goal of supporting PD patients’ diagnosis and remote monitoring.
Requirements: Knowledge of Matlab and Python
IoT smart ink pen for early detection and monitoring of patients with mild cognitive impairments and dementia
Supervisor: Prof. Simona Ferrante
Co-Supervisor: Simone Toffoli
Collaborations: Alice Naomi Preti Università degli Studi di Milano Bicocca
Aim: Dementia affects millions of people worldwide. Handwriting results from a complex network made up of cognitive, kinesthetic, and perceptual-motor abilities and it is one of the daily’s activities affected in patients with dementia. The aim of this work is to characterise the longitudinal evolution of handwriting in patients affected by dementia, using both an IoT smart ink and a digitizing tablet. Evaulated tasks include spontaneous handwriting (e.g., signature) and dictated handwriting.
Requirements: Basic knowledge of Matlab and Python
Teleconsultation to remotely assess learning abilities
Supervisor: Prof. Simona Ferrante
Co-Supervisor: Linda Greta Dui, Simone Toffoli
Collaborations: Università dell’Insubria
Aim:
Remote health has several advantages, from widening access to the healthcare service to reducing costs, and its importance became even more evident during the years of the pandemic, when access to hospitals was limited. However, the feasibility of performing clinical tests from a remote setting still needs to be verified. During the ESSENCE project, a teleconsultation app was developed and tested with children, that were evaluated for their reading and writing abilities, also by the use of an IoT smart ink pen.
The aim of the thesis is to further develop the application based on firts pilot tests already done and to test it in primary schools in collaboration with neuropsychologists.
Requirements: Interest in data analysis
Automation and enrichment of dysgraphia screening
Supervisor: Prof. Simona Ferrante
Co-Supervisor: Linda Greta Dui
Collaborations: Università dell’Insubria
Aim: Dysgraphia diagnosis is mainly based on writing speed on paper, but clinical test evaluation is subjective, time expensive, and the causes which underlie such slowness are rarely investigated in clinical practice. The aim of this thesis is to provide an autmatic scoring for dysgraphia tests staring from images, and to enrich it with data coming from an IoT smart ink pen used diring their execution
Requirements: Basic knowledge of Matlab, R or Python; interest in Deep Learning
Computational Neuroscience
Development of realistic mouse whiskers for a mouse neurorobot
Supervisor: Prof. Alberto Antonietti, Prof. Alessandra Pedrocchi
Collaborations: University of Pavia
Aim: The aim of the thesis is to enhance the existing robotic whisking system by incorporating more realistic properties into the design. The research focuses on a mouse robot model that currently consists of four rigid whiskers. The goal is to extend this model by introducing whiskers with elastic properties, thereby making them more similar to actual mouse whiskers. Additionally, the thesis aims to integrate these realistic whiskers into an active whisking task, where the robot can actively engage with its environment using its whiskers.
In summary, the main objectives of the thesis are as follows:
– Enhancing Realism: Modify the existing rigid whiskers to possess elastic properties, mirroring the flexibility and sensitivity of real mouse whiskers. This enhancement is crucial for the robot to accurately perceive and interact with its surroundings, replicating the way a mouse uses its whiskers for tactile sensing.
– Integration into an Active Whisking Task: Develop a control system that enables the robot to actively whisk and perform tasks based on the sensory input received from its whiskers. This involves programming the robot to move its whiskers purposefully, interpret the feedback obtained through whisker interactions, and respond accordingly.
– Sensorimotor Integration: Study the integration of sensory information from the whiskers with the motor control system of the robot. This involves understanding how the robot processes tactile data from its whiskers and translates it into appropriate motor responses, enabling it to navigate and interact with the environment effectively.
– Validation and Testing: Rigorously test the enhanced robotic whisking system to validate its effectiveness and reliability. This includes assessing its ability to accurately perceive different textures, distances, and objects through whisker interactions and evaluating its performance in various real-world scenarios.
By achieving these objectives, the thesis aims to contribute valuable insights into the development of advanced robotic systems that can mimic the sensory and motor capabilities of living organisms, specifically focusing on the tactile sensing abilities of mice. This research has the potential to advance the field of robotics, leading to the creation of more sophisticated and capable robots for various applications, including those in neuroscience, artificial intelligence, and human-robot interaction.
Requirements: Strong coding skills in Python and C++, passion for robotics and soft robotics topics, and attitude to independent work.
Integrating Machine Learning Algorithms for Enhanced Reproducibility in Biology Studies
Supervisor: Prof. Alberto Antonietti (alberto.antonietti@polimi.it)
External Co-Supervisors:
Dr. Tracey Weissgerber (tracey.weissgerber@bih-charite.de)
Vladislav Nachev, Data Scientist (vladislav.nachev@charite.de)
Collaborations: QUEST Center Berlin Institute of Health at Charité (BIH)
Description:
Aim: Western blotting, a widely utilized technique in molecular biology, serves as a fundamental method for detecting and quantifying protein expression levels. Despite its prevalence, concerns persist regarding the reproducibility and transparency of published western blot data. This thesis project aims to develop automated tools for the identification of problematic practices in western blot images, addressing the observed shortcomings in publication standards. By leveraging image processing and machine learning techniques, the project seeks to enhance the detection of issues such as narrow cropping, absence of molecular weight markers, and inadequate molecular weight labeling.
Project phases:
-Understanding issues of western blots and how to identify them in images
-Training in data curation of western blots to enrich the training database
-Identification of the techniques that best balance performance with efficiency
-Formulation and implementation of the proposed Machine Learning model
-Extensive testing of the developed tool
-Validation of the tool performances with new input data
Requirements:
– Willingness to deepen the knowledge in Meta-Research and Reproducibility topics
– Python programming skills (to be acquired before starting the project)
Other information:
Possibility to perform part of the work hosted at the QUEST Center in Berlin
Possibility of doing the thesis in pairs
Futher readings:
https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.3001783#abstract0
Development of Methods for Assessing Structural Connectivity from Diffusion Tensor Imaging
Supervisor: Prof. Emilia Ambrosini (emilia.ambrosini@polimi.it)Co-Cosupervisor: Dott. Alfonso Mastropietro
Collaborations: Istituto di Sistemi e Tecnologie Industriali Intelligenti per il Manifatturiero Avanzato (STIIMA) del Consiglio Nazionale delle Ricerche – Milano
Aim: to develop and evaluate methods for assessing structural connectivity from diffusion tensor imaging (DTI), a technique that measures the diffusion of water molecules in brain tissue. Structural connectivity refers to the anatomical connections between different brain regions, which can be inferred from the direction and magnitude of water diffusion along white matter tracts. DTI can provide valuable information about the integrity and organization of the brain’s wiring, which is relevant for various neurological and psychiatric disorders.
The project phases are: 1.Literature review: A comprehensive survey of the existing methods for structural connectivity analysis from DTI, including tractography, graph theory, network analysis, and machine learning approaches: 2. Method development: The design and implementation methods for structural connectivity assessment from DTI, based on the state-of-the-art techniques; 3. Method evaluation: The testing and validation of the proposed methods on real datasets, using appropriate metrics and benchmarks.
Requirements: Knowledge of Python, MATLAB, 3D-Slicer, ITK-SNAP, MrTrix
VADIS – Development of an easy-to-use computational model of Astrocytes for large Spiking Neural Networks
Supervisor: Prof. Alberto Antonietti (alberto.antonietti@polimi.it)Co-Cosupervisor: Carlo Andrea Sartori
Collaborations: MiMic Lab (Prof. Marco Rasponi, Paola Ochetta), TU Graz
Aim: In the context of the VADIS project (in Vitro AnD In Silico multi-organ-on-chip for functional evaluation and modeling of neuronal development and brain-environment interaction) we aim to develop an in silico twin model of a 3D organ-on-chip comprehending Neural, Glial and Vascular components. In this perspective, as an intermediate step, the aim of the Master Thesis would be the development of an Astrocyte model comprehending both the neuron-astrocyte (tripartite synapses) and the astrocyte-astrocyte interactions. In literature can be found plenty of models describing the exchange of signals between neuron and astrocytes. However, there are only a few models aiming at doing it in the NEST simulator context. NEST is a simulator used to model large-scale spiking neural networks (SNNs)with precise control of biophysical properties. For this project, the work will be implemented using both the NEST simulator and NESTML, where the astrocyte model will be built.
The project phases are: 1. Literature Review on the present computational models of Astrocytes 2. Method development: Integration of the models in a single and simple one, suitable for large SNN simulation; 3. Method evaluation: Validation and simulation of the model.
Requirements: Knowledge of Python; Knowledge of C++ (not mandatory, but preferable); Willingness to work in a collaborative environment
Master Thesis opportunities in theoretical neuroscience
Supervisor: Prof. Alberto Antonietti (alberto.antonietti@polimi.it)Co-Cosupervisor: Alessandro Sanzeni, Carlo Andrea Sartori
Collaborations: Bocconi University
Aim: Thesis opportunities are available at Alessandro Sanzeni’s group at Bocconi University in Milan. Their research tackles fundamental questions in neuroscience by integrating theoretical approaches—mathematical modeling, statistical physics, and machine learning—with experimental data. Currently, we focus on understanding how the visual system supports complex functions, such as object recognition and tracking. We approach this from multiple perspectives:
Functional models: Neuroanatomical evidence shows that the visual system in the brain consists of multiple interconnected areas arranged in a shallow hierarchy. How is visual information transformed along this hierarchy? Do different areas gradually transform raw sensory input into higher-level abstract representations, similar to modern deep learning models? Or do individual areas specialize in specific computations? To address these questions, we are developing and probing data-driven models of the visual hierarchy.
Mechanistic models: How do neural responses to visual stimuli emerge from the collective dynamics of neurons? We investigate this process at multiple scales—from dendritic computations at the single-cell level to neural interactions within local circuits and multi-area interactions at the macro scale. Using a physics-inspired approach, we employ simple models constrained by experimental data to identify the key mechanisms driving visual processing.
Normative models: Why are neural circuits structured as they are? Are features of brain networks different from artificial ones merely biological quirks, or do these features provide specific computational benefits? We aim to answer these questions using simplified models to explore the impact of particular features on computations, guided by a statistical physics approach.
Requirements: Knowledge of Python; Willingness to work in a collaborative environment
Modelling the mammalian brainstem to explore the neural mechanisms of sound localization
Supervisor: Prof. Alberto Antonietti
Co-Supervisor: Francesco De Santis
Collaborations: Imperial College London, Università di Genova
Aim: Implementing bioinspired neural networks in a computational environment is a formidable instrument to shed light into the complexity of brain processes. Such networks grant access to the dynamic activities of individual neurons within intricate circuitry, particularly those engaged in executing neurosensory functions.
This thesis proposes to study how the mammalian brainstem implements sound localization: the ability to identify an acoustic source in the surrounding space. We have reconstructed a model made of thousands of spiking neurons tailored to the auditory brainstem circuitry and its tonotopic organization.
The candidate will work hand by hand with supervisor and co-supervisor in attempt to improve the overall performance of the brainstem computational model, to replicate the experimental evidences obtained by several studies in mammals and to introduce realistic learning mechanisms in the network, such as spike timing dependent plasticity (STDP) and other synaptic processes concerning different physiological signaling molecules.
Requirements: knowledge of Python; curiosity for neuroscientific topics; willingness to work in a collaborative environment
Simulating a thalamic motor controller
Supervisor: Prof. Alberto Antonietti, Alessandra Pedrocchi
Co-Supervisor: Eng. Francesco Sheiban
Collaborations: University of Pavia
This master’s thesis focuses on the refinement and validation of a spiking neural network (SNN) model representing mouse motor thalamic nuclei. The network has been anatomically designed integrating data at multiple scales to place and connect neurons, yet it remains unexplored in terms of simulation outcomes.
The primary objective is to tune the model parameters systematically to align simulated neural dynamics with experimental electrophysiological data. Upon successful parameterisation, the model will serve as a powerful tool for investigating system controller properties and simulating behavioural tasks associated with mouse motor thalamic function.
Methodology:
Detailed examination of the existing spiking neural network model, dynamical systems analysis and characterisation, validation of the tuned model through rigorous comparison with empirical data. The thesis will be carried out using the NEST Simulator.
Requirements:
The student undertaking this project should possess strong Python programming skills and a foundational understanding of basic computational neuroscience concepts. While the spiking neural network has already been implemented, a grasp of dynamical systems theory is crucial for effective parameter tuning and interpretation of simulation outcomes.
Implementation of methods and analysis for cardiac excitable tissue on a multi electrode array system
Supervisor: Prof. Alessandra Pedrocchi, Prof. Alberto Antonietti, Dott. Andrea Menegon, Dott. Giovanni Peretto
Co-Supervisor: Francesco De Santis
Collaborations: IRCCS San Raffaele Hospital
Aim:
The use of MEA (Multi Electrode Array) is a well-accepted technique for recording electrical signals from excitable cells and tissues with high spatial and temporal resolution. The ability to record multiple evoked-responses in parallel at different electrodes within a single slice and within the same region of interest increases the reliability of the data and their subsequent statistical power. However, data from MEA recordings have traditionally been analysed manually, which is labour-intensive, slow and often user dependent.
This thesis proposes a comparative study on cardiac tissue affected by myocarditis, with particular emphasis to the development of an innovative software for the MEA readout. The work will be held at the IRCCS San Raffaele Hospital, in strict collaboration with the doctors, and it will cover each step of the experimental workflow, from the realization of in vitro cultures of cardiomyocytes to the design of the electronic set-up and the implementation of the the readout system.
Requirements: knowledge of Python; electronic background; willingness to develop different skills; interest in medicine and in pathologies
Signal encoding in a morphologically detailed cerebellar spiking model for simulating an adaptive reaching task
Supervisor: Prof. Alberto Antonietti, Alessandra Pedrocchi
Co-Supervisor: Eng. Benedetta Gambosi
Collaborations: University of Pavia
Aim: The bottom-up strategy is one possible approach to studying the brain. It attempts to construct biologically realistic neuronal network models based on the detailed knowledge of the constituting components. The assumption is that with sufficient accuracy in structure and dynamics, the correct functions will emerge. Toward this aim, we want to simulate a control system integrated with a detailed model of the cerebellum able to perform an adaptive reaching task.
The candidate will first investigate how to encode motion-related signals as cerebellar inputs; then, they will analyse how these signals propagate into the cerebellar circuit and turn into the physiological cerebellar output signals. The model’s overall performance will be evaluated by embedding it in a comprehensive control system to drive a closed-loop reaching task simulation.
Requirements: Knowledge of Python; Knowledge of C++ (not mandatory, but preferable); Willingness to work in a collaborative environment
