• Graduate Research Fellow at the National Research Council (CNR) in Pisa, Italy
• PhD student in Information Engineering at the University of Pisa. Project: “Boosting deep learning with causality: insights on medical imaging”.
• Some of my current interests span feature disentanglement, causal representation learning, and model robustness through causality.

• CMRxReconChallenge: A group project together with other PhD students to respond to the MICCAI 2023 challenge https://cmrxrecon.github.io/ regarding cardiac cine MR reconstruction. It was so nice to code all together! Additional contributions here and here. You can read our paper Cine cardiac MRI reconstruction using a convolutional recurrent network with refinement to discover how we performed in the MICCAI STACOM workshop 2023.
• A causal perspective on medical image classification via neural networks (causal_medimg): by building on ideas from this work, I am developing the code and expanding its capabilities with some novel architectural modifications. When it comes to medical image classification, assessing whether DL algorithms truly capture the cause-and-effect relationship between diseases and their underlying causes (or merely learn to map labels to images) remains a challenge. This work aims at disentangling features that causally influence the diagnosis and features that act as confounders. Work in progress.
• Boosting CNNs with knowledge of conditional asymmetries across feature maps (causality_conv_nets): Code for experimenting with causality-aware (driven) CNNs, where the network learns and exploits intrinsic information contained in image datasets regarding the causal disposition of object in the visual scene. See our pre-print Exploiting Causality Signals in Medical Images: A Pilot Study with Empirical Results where we introduce the architecture and the "causality factor extractor". To see how such "causality"-driven neural networks can boost performance and XAI explanations in low-data scenarios (Few-Shot), check our ICCV 2023 paper Causality-Driven One-Shot Learning for Prostate Cancer Grading from MRI!!
• (Un)conditional Denoising Diffusion Probabilistic Models: implementation of DDPMs in PyTorch for generating synthetic medical images of prostate cancer patients. I am developing both unconditional and conditional (classifier-free guidance) generation. Work in progress.
• ProtoPNet, with @andreaberti11235: we investigate the application of prototypical part learning to the medical imaging field. In particular, mammographic images are used. The clinical question is to determine the malignancy of breast masses, and the goal is to assess that using Deep Learning methods, without an invasive biopsy, which still remains the gold standard today
• dataset_utils_scripts, with @andreaberti11235: useful scripts for deep learning pipelines with digital images. Includes: stratified group splitting for dataset preparation, code for early stopping in neural networks' training, image resize and histogram of dimensions, scripts for reading DICOM/NIFTI files and convert them in PNG images, pipelines for Data Augmentation, etc

• ProCAncer-I European Union Project funded by Horizon 2020 research and innovation programme under grant agreement No 952159
• TAILOR European Union Project ICT-48 Network (GA 952215). A network of AI research excellence centre
• NAVIGATOR Tuscany Regional Project. An Imaging Biobank to Precisely Prevent and Predict cancer, and facilitate the Participation of oncologic patients to Diagnosis and Treatment
• PRAMA Tuscany Regional Project. Proteomics, RAdiomics & Machine learning-integrated strategy for precision medicine for Alzheimer’s

#DeepLearning #MachineLearning #ArtificialIntelligence #MedicalImaging #Causality #Explainability #GenerativeModels