Alessandra Retico
INFN
Sezione di Pisa
Artificial Intelligence in Medicine:
focus on Multi-Input Analysis
Update meeting referee - Preventivi CSN5 2026
Outline
Outline
A. Retico
AIM_MIA
Artificial Intelligence in Medicine: focus on Multi-Input Analysis
2
Sept. 12, 2025
Report 2025
Report 2025
Plans 2026
kick-off meeting in Pisa, February 13, 2025 https://agenda.infn.it/event/45571/
Main meeting outputs:
⇒ Refinement of the list of WP and task coordinators
⇒ Establishment of an operational working group on WP3 activity (data platform and computing infrastructure) with at least one collaborator per group
⇒ Identification of public datasets of interest to be made available on the common data platform and priority ranking
A. Retico
AIM_MIA
Artificial Intelligence in Medicine: focus on Multi-Input Analysis
3
Sept. 12, 2025
Report 2025: WP and task coordinators
Report 2025
Plans 2026
Updated Gantt with coordination roles
A. Retico
AIM_MIA
Artificial Intelligence in Medicine: focus on Multi-Input Analysis
4
Sept. 12, 2025
Report 2025: activity WP1 and WP2
Report 2025
Plans 2026
True goal: find research targets on which to collaborate by dividing tasks based on our peculiar expertise
A. Retico
AIM_MIA
Artificial Intelligence in Medicine: focus on Multi-Input Analysis
5
Sept. 12, 2025
WP1. Mining multi-modal information
Chair: F. Sensi (GE)
Task1.1 Feature-based approach to multi-input analysis
Task leader: P. Oliva (PI)
Task1.2 Integration of multi-parametric and multi-modal imaging data
Task leader: C. Talamonti (FI)
Task1.3 AI solutions for heterogeneous data analysis
Task leader: M. Marrale (CT)
| WP1 | | WP2 | | | |
(Multi-)input data | Method | Goal | Method | Goal | Clinical area | Group |
CT, PET | Radiomics + ML, XAI | improve diagnostic power | | | Lung Tumor | BA |
MRI T1, T2, clinical data | ML and feature selection (DnetPRO) | data compression | | | | BO |
CT, CT perfusion | image analysis | quantification | | | | BO |
DWI | | | GAN | improve angular resolution | | BO |
MRI, fMRI | Radiomics + ML, XAI | case-control classification | | | ASD | CA |
MRI T1 | Radiomics + ML, XAI | case-control classification | | | Alzheimer | CT |
MRI T1 | Unet | segmentation of brain structures | | | Parkinson | CT |
CT sinogram | | | VAE | data curation (e.g., artifacts) | | FE |
CT, PET, MRI, dose | Radiomics + ML, XAI | metastasis prediction | | | soft tissue sarcoma | FI |
FDG-PET, MRI, clinical data | Image processing, statistical analysis and ML | quantification | | | proteinopathiae | GE |
PT-tau, MRI, clinical data | Graph analysis, Random Forest | patient outcome | | | multiple sclerosis | GE |
DCE-MRI, clinical data | Radiomics + ML, XAI; federated learning | detection of TNBC | traditional techniques, | harmonization, missing data curation, sample balanching | breast cancer | LE |
PET/CT | MatRadiomics; Radiomics + ML | segmentation; classification | | | breast cancer | LNS |
CT, clinical data | Radiomics + ML, Lasso, XAI | response to therapy | | | hepatocellular carcinoma | MI |
MRI, T1 T2 FLAIR | Radiomics + ML, XAI; SAM | classification; segmentation | | | breast cancer | PI |
CT | 2D, 3D Unet | segmentation | | | pancreas tumor | PI |
CT | 2D, 3D Unet | segmentation | | | lung tumor | PI |
WP2. Handling incomplete/missing/limited datasets
Chair: C. Testa (BO)
Task2.1 Traditional approaches for data curation and augmentation
Task leader: G. De Nunzio (LE)
Task2.2 Medical Image Data Generation
Task leader: C. Testa (BO)
Task2.3 Data inpainting with CNN
Task leader: G. Di Domenico (FE)
Report 2025: activity WP3
Report 2025
Plans 2026
A. Retico
AIM_MIA
Artificial Intelligence in Medicine: focus on Multi-Input Analysis
6
Sept. 12, 2025
It will be based on an open-source imaging IT platform developed by Washington University for neuroimaging data analysis: XNAT
Thanks to its extensibility, XNAT can be used to support a wide range of imaging-based projects.
T3.1 Definition of requirements and user roles
T3.2 Realization of the data platform prototype
WP3. Data platform and computing infrastructure
Chair: F. Lizzi (PI)
Task3.1 Definition of requirements and user roles
Task leader: C. Scapicchio (PI)
Task3.2 Realization and maintenance of the data platform prototype
Task leader: A. Formuso (PI)
Task3.3 Integration of data processing pipelines and output storing
Task leader: F. Sensi (GE)
Task3.4 SW organization and repository
Task leader: I. Postuma (PV)
Task3.5 Data collection
Task leader: N. Curti (BO)
Report 2025: activity WP3
Report 2025
Plans 2026
A. Retico
AIM_MIA
Artificial Intelligence in Medicine: focus on Multi-Input Analysis
7
Sept. 12, 2025
T3.3 Integration of data processing pipelines and output storing (starts in July 2026)
T3.4 SW organization and repository
T3.5 Data collection
WP3. Data platform and computing infrastructure
Chair: F. Lizzi (PI)
Task3.1 Definition of requirements and user roles
Task leader: C. Scapicchio (PI)
Task3.2 Realization and maintenance of the data platform prototype
Task leader: A. Formuso (PI)
Task3.3 Integration of data processing pipelines and output storing
Task leader: F. Sensi (GE)
Task3.4 SW organization and repository
Task leader: I. Postuma (PV)
Task3.5 Data collection
Task leader: N. Curti (BO)
Report 2025: motivation and ideas for dataset selection
Report 2025
Plans 2026
Priority was given to these 4 datasets:
The upload of other datasets of interest will follow during 2026
A. Retico
AIM_MIA
Artificial Intelligence in Medicine: focus on Multi-Input Analysis
8
Sept. 12, 2025
Report 2025: activity WP4
Report 2025
Plans 2026
A. Retico
AIM_MIA
Artificial Intelligence in Medicine: focus on Multi-Input Analysis
9
Sept. 12, 2025
WP4. Project management, outreach and networking
Chair: A. Retico (PI)
Task4.1. Project management and networking
Task leader: A. Retico (PI)
Task4.2. Collaboration with AIFM and other associations
Task leaders: C. Talamonti (FI), A. Chincarini (GE);
Task4.3. Outreach
Task leader: ME Fantacci
WP4 | Project management, outreach and collaborations with external parties | Task coordinator(s) | Task participants | Activity |
| T4.1 Project management and networking (A. Retico) | A. Retico | local group coordinators and WP leaders | The board of group coordinators and WP chairs:
The group coordinators and WP chairs organize local meetings and WP meetings, respectively All participants to the AIM_MIA project are included in the aim-all@pi.infn.it |
| T4.2 Collaboration with AIFM and medical associations | C. Talamonti, A. Chincarini | all groups |
|
| T4.3 Outreach | M.E. Fantacci | all groups | All groups are involved in outreach activities (Scienzestate 2025, 11-12 June 2025, Firenze Bright Night 2025, Sept 26th, some sections) |
People & Budget 2025
Report 2025
Plans 2026
A. Retico
AIM_MIA
Artificial Intelligence in Medicine: focus on Multi-Input Analysis
10
Sept. 12, 2025
Richieste risorse di calcolo su infrastruttura INFN nazionale
People 2025
Budget 2025 (richiesto)
85 persone, 27.45 FTE (~0.3 FTE per persona)
| Sede | Miss | Miss SJ | Cons | Pub | Pub SJ | Manut | Invent | Lic SW | Tot | Tot SJ |
| BA | 3 | | 1 | | | | | | 4 | |
| BO | 3 | | 1 | | | | | | 4 | |
| CA | 5 | | 2 | | | | | 0.5 | 7.5 | |
| CT | 3 | | 2 | | | | | | 5 | |
| FE | 2 | | 2 | | | | | | 4 | |
| FI | 5 | | 2 | | | | | | 7 | |
| GE | 2 | 2 | 1.5 | | | 2 | | 0.5 | 6 | 2 |
| LE | 2 | | 0.5 | | | | | | 2.5 | |
| LNS | 2 | | 1 | | | | | | 3 | |
| MI | 2.5 | | 2.5 | | | | | | 5 | |
| PI | 7 | | 2 | | 10 | | | 0.5 | 9.5 | 10 |
| PV | 5 | | 2 | | | | 2 | | 9 | |
| Tot | 41.5 | 2 | 19.5 | | 10 | 2 | 2 | 1.5 | 66.5 | 12 |
Budget assegnato: 67.5 kE
Report 2025: Milestones 2025
Report 2025
Plans 2026
A. Retico
AIM_MIA
Artificial Intelligence in Medicine: focus on Multi-Input Analysis
11
Sept. 12, 2025
Deadline | Milestones | Achievements | Proposed % of completion | |
December 31, 2025 | M1.1 | Development of a DL-based analysis pipeline for feature extracted from multiple sources and release to the SW repository | Several analysis pipelines have been implemented for multi-input data (func, struct, blood, gene data) and are under testing. Maturity of each solution will be evaluated to find readiest candidate for the release in the SW repository | 40% |
December 31, 2025 | M2.1 | Implementation and release to the SW repository of traditional data augmentation and data imputation techniques. | Traditional data augmentation techniques have been implemented in many SW packages developed by the research group members. The integration on the SW repository has to be completed | 40% |
March 31, 2025 | M3.1a | Initialization of the shared SW repository and release of guidelines for contributors | 90% | |
December 31, 2025 | M3.1b | Release of an internal note reporting the guidelines to set up the data platform (definition of the requirements, user roles, allowed data types) | Set-up information have been collected (including requirements, user roles, allowed data types); they have to be organized in a harmonic report | 80% |
March 31, 2025 | M4.1 | Kick-off meeting organization and release of a plan of collaboration with external parties | kick off meeting (Febr 2025) https://agenda.infn.it/event/45571/ Plan of collaboration not released | 80% |
Papers 2025
Report 2025
Plans 2026
A. Retico
AIM_MIA
Artificial Intelligence in Medicine: focus on Multi-Input Analysis
12
Sept. 12, 2025
Berti, Andrea et al. | An explainable-by-design end-to-end AI framework based on prototypical part learning for lesion detection and classification in Digital Breast Tomosynthesis images | COMPUTATIONAL AND STRUCTURAL BIOTECHNOLOGY JOURNAL | 2025 |
Biondi, Riccardo et al. | Towards Precision Medicine in Sinonasal Tumors: Low-Dimensional Radiomic Signature Extraction from MRI | DIAGNOSTICS | 2025 |
Buzzi, Simone et al. | A Time-Series Approach for Machine Learning-Based Patient-Specific Quality Assurance of Radiosurgery Plans | BIOENGINEERING-BASEL | |
Cabini, Raffaella Fiamma et al. | A KINETIC APPROACH TO CONSENSUS-BASED SEGMENTATION OF BIOMEDICAL IMAGES | KINETIC AND RELATED MODELS | 2025 |
Damiani, Simone et al. | Deep Learning Denoising of Low-Dose Computed Tomography usingConvolutional Autoencoder: a Phantom Study | PROCEEDINGS OF THE 12th CONFERENCE ON BIOIMAGING | 2025 |
Lossano, Simone et al. | Generative super-resolution AI accelerates nanoscale analysis of cells | MACHINE LEARNING-SCIENCE AND TECHNOLOGY | 2025 |
Marini, Lorenzo et al. | Artificial intelligence for in-vivo dosimetry using EPID in external beam photon radiotherapy | RADIOTHERAPY AND ONCOLOGY | 2025 |
Novielli, Pierfrancesco et al. | Personalized colorectal cancer risk assessment through explainable AI and Gut microbiome profiling | GUT MICROBES | 2025 |
Pelagotti, Virginia et al. | [18F]FDG PET metabolic correlates of cerebrospinal fluid TAR DNA-binding protein 43 in prodromal Alzheimer's disease | NEUROBIOLOGY OF DISEASE | 2025 |
Ricchi, Mattia et al. | Connectivity related to major brain functions in Alzheimer disease progression: microstructural properties of the cingulum bundle and its subdivision using diffusion-weighted MRI | EUROPEAN RADIOLOGY EXPERIMENTAL | 2025 |
Ricchi, Mattia et al. | Assessment of a diffusion phantom for quality assurance in brain microstructure diffusion MRI studies | SCIENTIFIC REPORTS | 2025 |
Tenerani, Maria Irene et al. | Use of radiomics in low dose chest CT: a proposal for a phantom multi-centric study | PROCEEDINGS OF THE 12th CONFERENCE ON BIOIMAGING | 2025 |
Tiddia, Gianmarco et al. | Explainable AI Highlights the Most Relevant Gait Features for Neurodegenerative Disease Classification | APPLIED SCIENCES-BASEL | 2025 |
Tsewalo Tondji Idriss Cabrel et al. | 2.5D Deep Learning Model with Attention Mechanism for Pancreas Segmentation on CT Scans | Proc. 18th International Joint Conference on Biomedical Engineering Systems and Technologies | 2025 |
Theses 2025
Report 2025
Plans 2026
Tesi Triennali
Matteo Bergamaschi (UniMI), Analisi di trattamenti chemioterapici di lesioni al fegato con tecniche di machine learning
Mounia FELALI BELGHITI (UniFI), L'intelligenza artificiale in medicina: le nuove responsabilità del tecnico sanitario di radiologia medica
Tesi Magistrali
Simone Santoro, (UniBO), Development of new approaches for generating synthetic Diffusion Gradient Directions in Diffusion Tensor Imaging
Giuseppe Antonio Motisi (UniPI), Human Connectome Analysis with Graph Neural Networks: the Autism Spectrum Disorder case study
Simone Damiani (UniPI), Denoising of Chest CT (Computed Tomography) images with a U-Net based Convolutional Autoencoder
Gianluca Morcaldi (UniPI), A machine learning approach to discriminate malignant and benign breast lesions using multimodal MRI
Martina Scagni (UniPV), Pre-processing, development, and evaluation of a convolutional neural network for diagnosing paranasal sinus diseases
Rosanna Tripi (UniPA, LNS), SEGMENTAZIONE DI IMMAGINI PRE-CLINICHE E CLINICHE ATTRAVERSO L’UTILIZZO DI DEEP NEURAL NETWORKS
Eleonora Giovagnoli (UniRM1, LNS), Automated Classification of Breast Tumors by a Radiomic Approach with matRadiomics toolbox to Distinguish Benign from Malignant Lesions
Lorenzo Piana (UniGE) Sviluppo di un modello dinamico a compartimenti per la valutazione delle malattie dei giunti in pazienti pediatrici
Tesi di Dottorato
Andrea Berti (UniPI), Towards Trustworthy AI in Healthcare: Transparent and Interpretable Models for Medical Imaging
Carlo Golini (UniBO), CT-based radiomics for differentiating post trombectomy hyperdensities
Matteo Parodi (UniGE) Generazione di un sistema di coordinate "clinically-driven" per la valutazione delle proteinopatie in dati multimodali
A. Retico
AIM_MIA
Artificial Intelligence in Medicine: focus on Multi-Input Analysis
13
Sept. 12, 2025
DEADLINE
PREMIO RESMINI
PREMIO VITA FINZI
Plans 2026: WP1 and WP2
Report 2025
Plans 2026
Several methods developed in 2025 can be generalized to work on datasets identified as of common scientific interests by different research groups
A. Retico
AIM_MIA
Artificial Intelligence in Medicine: focus on Multi-Input Analysis
14
Sept. 12, 2025
WP1. Mining multi-modal information
Chair: F. Sensi (GE)
Task1.1 Feature-based approach to multi-input analysis
Task leader: P. Oliva (PI)
Task1.2 Integration of multi-parametric and multi-modal imaging data
Task leader: C. Talamonti (FI)
Task1.3 AI solutions for heterogeneous data analysis
Task leader: M. Marrale (CT)
| WP1 | | WP2 | | | |
(Multi-)input data | Method | Goal | Method | Goal | Clinical area | Group |
CT, PET | Complex networks | patients' stratification | Lung Tumor | | | BA |
3D PET | | | GenAI | Inprove spatial resolution, case-control classification | syntetic data | BO |
MRI, fMRI | Radiomics + ML, XAI | case-control classification | | | Clinical areas TBD | CA |
MRI, fMRI, clinical data | Radiomics + ML, XAI; CNN | case-control classification | | | Clinical areas TBD | CT |
CT sinogram | | | VAE | data curation (e.g., missing angles) | | FE |
CT, PET, MRI, dose | Radiomics + ML, XAI | metastasis prediction | | | soft tissue sarcoma | FI |
DCE-MRI, clinical data | Radiomics + ML, XAI; federated learning | detection of TNBC | GenAI | curation of missing clinical data | breast cancer | LE |
PET/CT, clinica data | Correlation between imaging and clinical data | predicton of patient's outcome; classification | | | breast cancer | LNS |
CT, clinical data | Radiomics + ML, Lasso, XAI | response to therapy | standard techniques | data augmentation | hepatocellular carcinoma | MI |
CT, MRI | Unet; Radiomics + ML | segmentation, prediction patient's outcome | | | pancreas tumor | PI |
micro-CT, CT | Radiomics | virtual biopsy | | | lung tumor | PV |
WP2. Handling incomplete/missing/limited datasets
Chair: C. Testa (BO)
Task2.1 Traditional approaches for data curation and augmentation
Task leader: G. De Nunzio (LE)
Task2.2 Medical Image Data Generation
Task leader: C. Testa (BO)
Task2.3 Data inpainting with CNN
Task leader: G. Di Domenico (FE)
Plans 2026: WP3
Report 2025
Plans 2026
[T3.2] Definition of the final structure of the IT platform.
[T3.3] Development of a first strategy to integrate data processing pipelines (e.g. image pre-processing) to be executed directly from the platform and to store back the output.
[T3.4] Improve the quality of the SW repository and increment its population.
[T3.5] Upload of the public datasets of interest by the different participants.
A. Retico
AIM_MIA
Artificial Intelligence in Medicine: focus on Multi-Input Analysis
15
Sept. 12, 2025
Plans 2026: WP4
Report 2025
Plans 2026
Project management
Collaborations
Outreach
New projects
A. Retico
AIM_MIA
Artificial Intelligence in Medicine: focus on Multi-Input Analysis
16
Sept. 12, 2025
People & Budget 2026
Report 2025
Plans 2026
A. Retico
AIM_MIA
Artificial Intelligence in Medicine: focus on Multi-Input Analysis
17
Sept. 12, 2025
People 2026
Budget 2026 total request: 80 kE
| Sede | Miss | Miss SJ | Cons | Sem | Sem SJ | Pub | Pub SJ | Invent | Tot | Tot SJ |
| BA | 3 | | 1 | | | | | | 4 | |
| BO | 2 | 1 | 1 | | | | | 1 | 4 | 1 |
| CA | 2 | | | | | | | | 2 | |
| CT | 3 | | 2 | | | | | | 5 | |
| FE | 2 | | 1 | | | | | | 3 | |
| FI | 3 | | 1 | | 5 | | | | 4 | 5 |
| GE | 2 | | 2 | | | | | | 4 | |
| LE | 2 | | 0.5 | | | | | | 2.5 | |
| LNS | 2 | | 1 | | | | | 3.5 | 6.5 | |
| MI | 2.5 | | 2 | | | | | | 4.5 | |
| NA | 1.5 | | 0.5 | | | | | | 2 | |
| PI | 7 | 3 | 2 | | | | 10 | | 9 | 13 |
| PV | 4 | | 1.5 | | | | | 5 | 10.5 | |
| Tot | 36 | 4 | 15.5 | | 5 | | 10 | 9.5 | 61 | 19 |
89 persone, 30.1 FTE (~0.3 FTE per persona)
Request of National computing resources
GPU: 9000 GPU-hours on GPU A100
(> 40GB di VRAM) for AI model training
Organization of webinars on AI with AIFM
Publication fees (SJ) concentrated in PI
Budget 2026: supporting documents
Report 2025
Plans 2026
A. Retico
AIM_MIA
Artificial Intelligence in Medicine: focus on Multi-Input Analysis
18
Sept. 12, 2025
HW requests
BO: 1 kE - Monitor per conferenze e meeting
LNS: 3.5 kE - Computer portatile Mac per installare matRadiomics (SW sviluppato dal gruppo) e realizzare analisi di radiomica e ML su dati PET in collaborazione con i medici.
PV: 5 kE - Acquisto workstation, reso necessario dal fatto che verranno acquisiti dati localmente senza la possibilita' di utilizzare facility nazionali di calcolo causa riservatezza dei dati, con particolare riferimento a dati preclinici MRI e CT, che richiedono grossa capacita' di memoria (anche GPU) e di calcolo rapido.
Budget 2026: assigned
Report 2025
Plans 2026
A. Retico
AIM_MIA
Artificial Intelligence in Medicine: focus on Multi-Input Analysis
19
Sept. 12, 2025
Plans 2026: proposed Milestones
Report 2025
Plans 2026
A. Retico
AIM_MIA
Artificial Intelligence in Medicine: focus on Multi-Input Analysis
20
Sept. 12, 2025
Deadline | Milestones | |
December 31, 2026 | M1.2 | Definition of a generalizable AI-based pipeline that integrated multiparametric or multimodal images |
December 31, 2026 | M2.2 | Inpainting of metal artifact corrupted sinograms with NN. |
December 31, 2026 | M3.2 | Instantiation of a XNAT platform prototype and integration of data sets |
June 30, 2026 | M4.2 | Identification of joint research or dissemination activities to promote the networking among researchers from different Institutions and associations |
Expected impact of AIM_MIA:
To be reformulated making them more detailed and verifiable
Plans 2026: Milestones
Report 2025
Plans 2026
A. Retico
AIM_MIA
Artificial Intelligence in Medicine: focus on Multi-Input Analysis
21
Sept. 12, 2025
Deadline | Milestones | |
June 30, 2026 | M1.2.a | Identification of at least two scientific use cases requiring AI-based multi-input analysis approaches and assignment of tasks among developers from different sites |
June 30, 2026 | M2.2.a | Integration on the SW repository of prototype versions of the proposed data curation techniques |
June 30, 2026 | M3.2.a | Instantiation of a XNAT platform and integration of the main datasets of interest for collaborative analyses |
December 31, 2026 | M1.2.b | Report on the preliminary results obtained by the multi-input analysis approaches implemented in the two scientific cases of common interest |
December 31, 2026 | M2.2.b | Evaluation of the performance of the DL algorithms for sinogram curation and for increasing the DWI angular resolution |
December 31, 2026 | M3.2.b | Release of the user manual for the integration of analysis pipelines into the XNAT platform |
December 31, 2026 | M4.2 | Organization of a webinar series on AI impact in Medicine in collaboration with AIFM |
��
�