arXiv subjects

Omar Nusrat

Publications and source records attributed to Omar Nusrat.

3 recordsLinked to original sources

arXiv2026

MedPhysBench: A Benchmark for Evaluating Language Models on Medical Physics Calculations and Error Detection

Medical physicists turn a prescription into a delivered dose and check one another's calculations, because one error in a beam calibration or a monitor-unit (MU) calculation can affect every patient treated with it. MedPhysBench has ten seeded generators of calculation items in radiation therapy, diagnostic imaging, nuclear medicine and radiation protection, up to clinical tasks such as a TG-51 calibration worksheet. Reference answers come from code implementing the governing protocols, and a grader converts units and applies tolerance bands. From the same inputs, separate checking items test error detection, showing a colleague's worksheet or only the colleague's result, half correct and half with one planted error. On a held-out test split of 1,200 items (half for two models), 12 models scored 33.3 to 100.0% on calculation. Of the nine answers graded incorrect for the four best (99.3-100.0%), five contained model errors and four followed NCRP Report 151 on points our items misstate or leave open. Below them, most models did worst on level-3 clinical tasks, and Claude Sonnet 5.5 omitted the inverse-square correction in 19 of 20 isocentric MU items. Five models detected at least 96.7% of mostly large planted errors with at most 4% false alarms; for four of the other seven, checking only the result raised false alarms significantly. In an exploratory targeted test with 20 public cases of an omitted inverse-square correction, the three best models tested caught it in every worksheet, and three weaker models accepted 16 to 18 of 20 planted worksheets, one of them while applying the correction in most of its own calculations. Claude Opus 5.5 accepted 6 of 20 wrong results, although it applied the correction in all 20 of its own. We report the effect of two problems in our shielding items, found after the runs, and release the code, per-item outcomes and targeted-test responses.

physics.med-ph↗

arXiv2026

Weight-space programming for auditing clinical decision networks

Machine learning models in radiotherapy are commissioned by testing them on cases with known answers, which may not reveal a shortcut that the test cases share with the training data. Frontier language models have recently been reported to read and write the raw weights of small neural networks, an ability called weight-space programming. The purpose of this study was to determine whether such a model, given only a clinical protocol and a network's weights, can audit and repair the network. Networks with 433 to 4,801 parameters were trained to screen head-and-neck planning requests against four rules. Flaws were planted through the training data: confounds that hid each rule, a 3 Gy shift in the cord threshold, and a hidden trigger. Claude Opus 5.5 audited each network blind, without tools, data or code execution, and every claim and repair was tested on about 420,000 requests. The language model named the broken rule in 16 of 18 flawed networks (95% CI 67% to 97%), including all 15 with a confound or a threshold shift. It repaired each of the 16 by changing a median of 3 parameters. Fine-tuning on protocol-labeled requests also repaired every network, but changed nearly every parameter. Each confound and shift it found was visible in single units, and a simple statistic chosen for each flaw type found the same flaws. It cleared 2 of 5 healthy networks without being forced and found 1 of 3 triggers. Testing against the protocol detected every flaw, and black-box testing found the confounds and the shift faster. The results indicate that, in small networks, weight-space programming can locate and repair flaws that silence an input or shift a threshold, but cannot yet certify that a network is correct.

physics.med-ph↗

arXiv2025

When AI Does Science: Evaluating the Autonomous AI Scientist KOSMOS in Radiation Biology

Agentic AI "scientists" now use language models to search the literature, run analyses, and generate hypotheses. We evaluate KOSMOS, an autonomous AI scientist, on three problems in radiation biology using simple random-gene null benchmarks. Hypothesis 1: baseline DNA damage response (DDR) capacity across cell lines predicts the p53 transcriptional response after irradiation (GSE30240). Hypothesis 2: baseline expression of OGT and CDO1 predicts the strength of repressed and induced radiation-response modules in breast cancer cells (GSE59732). Hypothesis 3: a 12-gene expression signature predicts biochemical recurrence-free survival after prostate radiotherapy plus androgen deprivation therapy (GSE116918). The DDR-p53 hypothesis was not supported: DDR score and p53 response were weakly negatively correlated (Spearman rho = -0.40, p = 0.76), indistinguishable from random five-gene scores. OGT showed only a weak association (r = 0.23, p = 0.34), whereas CDO1 was a clear outlier (r = 0.70, empirical p = 0.0039). The 12-gene signature achieved a concordance index of 0.61 (p = 0.017) but a non-unique effect size. Overall, KOSMOS produced one well-supported discovery, one plausible but uncertain result, and one false hypothesis, illustrating that AI scientists can generate useful ideas but require rigorous auditing against appropriate null models.

cs.AI↗