arXiv subjects

Adam Murphy

Publications and source records attributed to Adam Murphy.

3 recordsLinked to original sources

arXiv2026

Align then Refine: Text-Guided 3D Prostate Lesion Segmentation

Automated 3D segmentation of prostate lesions from biparametric MRI (bp-MRI) is essential for reliable algorithmic analysis, but achieving high precision remains challenging. Volumetric methods must combine multiple modalities while ensuring anatomical consistency, but current models struggle to integrate cross-modal information reliably. While vision-language models (VLMs) are replacing the currently used architectural designs, they still lack the fine-grained, lesion-level semantics required for effective localized guidance. To address these limitations, we propose a new multi-encoder U-Net architecture incorporating three key innovations: (1) an alignment loss that enhances foreground text-image similarity to inject lesion semantics; (2) a heatmap loss that calibrates the similarity map and suppresses spurious background activations; and (3) a final-stage, confidence-gated multi-head cross-attention refiner that performs localized boundary edits in high-confidence regions. A phase-scheduled training regime stabilizes the optimization of these components. Our method consistently outperforms prior approaches, establishing a new state-of-the-art on the PI-CAI dataset through enhanced multi-modal fusion and localized text guidance. Our code is available at https://github.com/NUBagciLab/Prostate-Lesion-Segmentation.

cs.CV↗

arXiv2022

Completing RHINO

The right-handed (RH) Higgs-induced neutrino mixing (RHINO) model explains neutrino masses and origin of matter in the universe within a unified picture. The mixing, effectively described by a dimension five operator, is responsible both for the production of dark neutrinos, converting a small fraction of seesaw neutrinos acting as source, and for their decays. We show that including the production of source neutrinos from Higgs portal interactions, their abundance can thermalise prior to the onset of source-dark neutrino oscillations, resulting into an enhanced production of dark neutrinos that thus can play the role of decaying dark matter (DM) for a much higher seesaw scale. This can be above the sphaleron freeze-out temperature and as high as $\sim 100\,{\rm TeV}$, so that strong thermal resonant leptogenesis for the generation of the matter-antimatter asymmetry is viable. We obtain a $\sim 1\,{\rm TeV}$--$1\,{\rm PeV}$ allowed dark neutrino mass range. Intriguingly, their decays can also explain a neutrino flux excess at ${\cal O}(100\,{\rm TeV})$ energies recently confirmed by the IceCube collaboration analysing 7.5yr HESE data. Our results also point to an effective scale for Higgs portal interactions nicely identifiable with the grandunified scale and many orders of magnitude below the effective scale for the mixing. We explain this hierarchy in a UV-complete model with a very heavy fermion as mediator: the first scale corresponds to the fundamental scale of new physics, while the second is much higher because of a very small coupling identifiable with a symmetry breaking parameter. Therefore, RHINO realises a simple unified model of neutrino masses and origin of matter in the universe currently under scrutiny at neutrino telescopes and potentially embeddable within a grandunified model.

hep-ph↗

arXiv2013

Genome-wide scan of 29,141 African Americans finds no evidence of selection since admixture

We scanned through the genomes of 29,141 African Americans, searching for loci where the average proportion of African ancestry deviates significantly from the genome-wide average. We failed to find any genome-wide significant deviations, and conclude that any selection in African Americans since admixture is sufficiently weak that it falls below the threshold of our power to detect it using a large sample size. These results stand in contrast to the findings of a recent study of selection in African Americans. That study, which had 15 times fewer samples, reported six loci with significant deviations. We show that the discrepancy is likely due to insufficient correction for multiple hypothesis testing in the previous study. The same study reported 14 loci that showed greater population differentiation between African Americans and Nigerian Yoruba than would be expected in the absence of natural selection. Four such loci were previously shown to be genome-wide significant and likely to be affected by selection, but we show that most of the 10 additional loci are likely to be false positives. Additionally, the most parsimonious explanation for the loci that have significant evidence of unusual differentiation in frequency between Nigerians and Africans Americans is selection in Africa prior to their forced migration to the Americas.

q-bio.PE↗