Skip to content
Drix10 Blog

Quantum model 91.9% accuracy in lung cancer prediction

, 8 items in AI in Healthcare and Science, 4 min read

In this digest (8 items)

Quantum model achieved 91.9% accuracy in preliminary lung cancer prediction tests using 108 training samples from an augmented dataset. Clinical validation is still pending.

Key points

  • Accuracy: 91.9% on preliminary tests.

  • Training data: 108 samples from an augmented dataset.

Sources

OpenMed GLiNER2.5 ClinicalNER Disease Spanish fine-tuned, F1 39.1-73.3

OpenMed GLiNER2.5 ClinicalNER Disease Spanish v1 is an open model for disease extraction in Spanish clinical text. After fine-tuning it reaches an F1 range of 39.1 to 73.3. It is released under the Apache-2.0 license.

Key points

  • F1 score after fine-tuning: 39.1 – 73.3

  • License: Apache-2.0

  • Detects four disease mentions (HTA, DM2, EPOC, IC) at exact character offsets

  • Model size: 0.3B parameters, runs on CPU.

Sources

Calcium treatments 19% larger stent areas vs shockwave

Calcium treatments produced about 19% larger coronary stent areas compared to shockwave treatment alone in patients with calcified nodules. Longer-term benefits are unproven.

Key points

  • 19% larger coronary stent areas with combined calcium treatments vs shockwave

  • Longer-term benefits remain unproven

Sources

Spendrule seed round raises $4M from Radical and ZealVC

Spendrule announced a $4 million seed round backed by Radical Investments and ZealVC. The AI platform checks health system invoices against contracts and flags overcharges.

Key points

  • Funding: $4M seed round

  • Investors: Radical Investments (including @mcuban) and ZealVC

Sources

Vorelios launch – research lab building AI that engineers anything

Vorelios announced its launch as a YC F26 research lab building AI that can engineer anything. The lab claims its models predict design behavior in seconds and use those predictions to invent and improve designs autonomously.

Key points

  • Launch: Vorelios is now live as a YC F26 company.

  • Claim: Models predict a design’s behavior in seconds and use predictions to autonomously invent and improve designs.

Sources

Glucose-sensing component cost one-tenth, performance comparable

Glucose-sensing component delivers comparable performance at roughly one-tenth the cost of an earlier alternative. The injectable sensor is still a proof of concept.

Key points

  • Performance: comparable to earlier alternative

  • Cost: roughly one-tenth the cost of earlier alternative

  • Status: injectable sensor remains a proof of concept

Sources

insitro's Daphne Koller appears on The Long Run podcast

insitro's Daphne Koller appears on The Long Run podcast as the latest guest. The episode is sponsored by AlphaSenseInc and Dash Bio.

Key points

  • Guest: Daphne Koller of insitro

  • Sponsors: AlphaSenseInc and Dash Bio

Sources

DoseCheck™ added to Doximity Ask dosing tool

DoseCheck™ is now available in Doximity Ask, giving physicians a built-in dosing check with source references. No language model is used, so there is no hallucination.

Key points

  • Feature: built-in check on dosing in Doximity Ask

  • Claim: no language model is involved, so there is no hallucination

Sources