arXiv subjects

Nicolò Savioli

Publications and source records attributed to Nicolò Savioli.

5 recordsLinked to original sources

arXiv2026

Gallileo-4D: Frozen Backbone Ensemble for Dynamic 4D Reconstruction

We describe our entry to the PhysAI Dynamic 4D Reconstruction Challenge, which placed third of 27 teams at 0.58356 APD on the final leaderboard, without a single gradient update. This was not the plan: of thirteen fine-tuning configurations of a pre-trained 4D backbone, twelve degraded the challenge score, and eleven of those twelve improved local validation at the same time. We trace this inversion to the structure of the benchmark: only 25% of the evaluation set belongs to the data variant released for training, so updates that fit the available data damage the pre-trained features the remaining 75% relies on. Our system therefore freezes the backbone and spends its budget at inference time, fusing three decoding configurations -- temporal stride-3, horizontal-flip test-time augmentation, and dense stride-1 -- under a convex weighting. The ensemble recovers +0.041 APD over the frozen baseline, more than any training run achieved, at zero training cost.

cs.CV↗

arXiv2026

TopoPult-SSL: Gland-Mask-Free Cross-Device Meibomian Gland Segmentation via Self-Distilled Weak Clinical Priors

Every new clinical imaging device creates a domain shift where dense gland masks are expensive yet cheap clinical signals -- eyelid outlines, Pult grades, morphometric ratios -- are routinely recorded. We present TopoPult-SSL, a two-stage framework for cross-device meibomian gland segmentation. Stage 1 adapts a source-trained model without target gland masks in the training loss, using four weak-prior anchors driven by target eyelid masks and clinical metadata only. Stage 2, when target gland masks are available, distils complementary Stage-1 teachers into a single compact student via supervised self-distillation. We develop and validate the technique on the public MGD-1k to CAMG research benchmark (1,000 to 100 images, different device), where the distilled model achieves Dice 0.716+/-0.006 (best 0.726), surpassing UA-MT (0.710) and the ensemble teacher (0.720) -- with a single pass. The gland-mask-free Stage-1 variant reaches Precision 0.694 vs. 0.30-0.34 for SAM/MedSAM (p<0.001), enabling deployment without dense gland contouring. Code and reproducibility scripts are released.

cs.CV↗

arXiv2026

Venice-H1: Failure-Aware Query Re-Ranking with Multi-Scale Grid Signatures for Referring Image Segmentation

Modern Referring Image Segmentation (RIS) systems generate multiple candidate masks per expression but rely on a simple heuristic--typically the argmax detection score--to select the final output. We identify query selection as a failure-case bottleneck: although heuristic selection succeeds on 82-93% of samples, the residual 7-18% of failures dominate the error budget, leaving a best-query selection gap of 3-11% mIoU. We introduce Venice-H1, a lightweight, backbone-decoupled post-hoc re-ranking module that encodes each candidate through multi-scale grid signatures--compact spatial descriptors pooled onto 4x4, 8x8, and 16x16 grids--and feeds them to a Transformer-based re-ranker with a Failure Gate (ROCAUC 0.78-0.82) that intervenes only when the default choice is likely suboptimal. Instantiated on DeRIS-L and DeRIS-B, Venice-H1 achieves delta_fail of +1.40 and +0.89 mIoU with strictly positive 95% CIs on all 16/16 (split, backbone) pairs and harmful-switch rates below 0.53%. Zero-shot transfer to medical referring segmentation (MS-CXR, M3D-RefSeg-2D) yields +1.16 and +0.51 mIoU without RIS-backbone fine-tuning. The module adds approximately 11.3M parameters and under 1 ms latency.

cs.CV↗

arXiv2020

One-shot screening of potential peptide ligands on HR1 domain in COVID-19 glycosylated spike (S) protein with deep siamese network

The novel coronavirus (2019-nCoV) has been declared to be a new international health emergence and no specific drug has been yet identified. Several methods are currently being evaluated such as protease and glycosylated spike (S) protein inhibitors, that outlines the main fusion site among coronavirus and host cells. Notwithstanding, the Heptad Repeat 1 (HR1) domain on the glycosylated spike (S) protein is the region with less mutability and then the most encouraging target for new inhibitors drugs.The novelty of the proposed approach, compared to others, lies in a precise training of a deep neural network toward the 2019-nCoV virus. Where a Siamese Neural Network (SNN) has been trained to distingue the whole 2019-nCoV protein sequence amongst two different viruses family such as HIV-1 and Ebola. In this way, the present deep learning system has precise knowledge of peptide linkage among 2019-nCoV protein structure and differently, of other works, is not trivially trained on public datasets that have not been provided any ligand-peptide information for 2019-nCoV. Suddenly, the SNN shows a sensitivity of $83\%$ of peptide affinity classification, where $3027$ peptides on SATPdb bank have been tested towards the specific region HR1 of 2019-nCoV exhibiting an affinity of $93\%$ for the peptidyl-prolyl cis-trans isomerase (PPIase) peptide. This affinity between PPIase and HR1 can open new horizons of research since several scientific papers have already shown that CsA immunosuppression drug, a main inhibitor of PPIase, suppress the reproduction of different CoV virus included SARS-CoV and MERS-CoV. Finally, to ensure the scientific reproducibility, code and data have been made public at the following link: https://github.com/bionick87/2019-nCoV

q-bio.QM↗

arXiv2019

Joint analysis of clinical risk factors and 4D cardiac motion for survival prediction using a hybrid deep learning network

In this work, a novel approach is proposed for joint analysis of high dimensional time-resolved cardiac motion features obtained from segmented cardiac MRI and low dimensional clinical risk factors to improve survival prediction in heart failure. Different methods are evaluated to find the optimal way to insert conventional covariates into deep prediction networks. Correlation analysis between autoencoder latent codes and covariate features is used to examine how these predictors interact. We believe that similar approaches could also be used to introduce knowledge of genetic variants to such survival networks to improve outcome prediction by jointly analysing cardiac motion traits with inheritable risk factors.

q-bio.QM↗