- Graduate research fellow at the National Research Council (CNR) in Pisa, Italy (Sep 2021 - Ongoing)
- Last-year PhD student in Information Engineering at the University of Pisa (Nov 2021 - Ongoing). Project: “Boosting deep learning with causality: insights on medical imaging”.
- My current interests and expertise span out-of-distribution robustness, domain generalization, and explainability through causal reasoning, feature disentanglement and contrastive learning.
- Proposed methods based on causal inference to foster deep learning models’ robustness to domain shift bias (out-of-distribution and domain generalization) via causal/spurious feature disentanglement, contrastive learning modules, and injection of prior knowledge. Applied to the classification of lung anomalies from large real-word-data chest X-ray datasets.
- Investigated techniques to discover causal disposition signals in images via Attention-inspired convolutional neural networks for cancer prediction from medical imaging (prostate MRI and H&E-stained digital pathology).
- Proposed biologically inspired context-aware image recognition networks akin to human vision.
- Experienced with DDPMs (generative AI) to synthesize MRI images of prostate cancer.
- Led a systematic literature review to study the intersection of causality and Explainable AI.
- Explored the applicability of prototypical part learning in medical imaging by experimenting with ProtoPNet on a breast masses classification from mammogram images.
- Conducted multivar analyses on clinical, biophysical, proteomic and optical data for Alzheimer's disease.
Machine learning & Deep learning | Computer Vision | Medical Image Analysis | Distributed computing
Programming: Python, Linux shell scripting & Bash, MATLAB, Java, Android, SQL.
Python libraries: PyTorch, CUDA, Scikit-learn, Pandas, NumPy, OpenCV, Biopython, PyDicom.
Tools: Git version control, Docker, SLURM-based large-scale HPC, NVIDIA DGX, DICOM/NIFTI formats.
You can find my published research on my Google Scholar profile.
- "OOD/DG robustness" Causality aids RObustness via COntrastive DIsentangled LEarning (CROCODILE): In this MICCAI 2024 paper at the UNSURE int. work., we propose a new deep learning framework to tackle domain shift bias on medical image classifiers and improve their out-of-distribution (OOD) performance, fostering domain generalization (DG). We showed how tools from causality can foster a model’s robustness via feature disentanglement, contrastive learning losses, and the injection of prior knowledge. Explore our project and read our paper to know more.
- "Biologically-inspired Computer Vision" CoCoReco: a connectivity inspired neural network for context aware recognition. Accepted at ECCV 2024's Human-inspired Computer Vision international workshop.
- "MRI Reconstruction" 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.
- "Causality/XAI" 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 ESWA 2024 paper 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!!
- "Generative AI" (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.
- "Explainable AI (XAI)" ProtoPNet, with @andreaberti11235: In our ICPR 2022 paper, 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