View PDF ↗
PDF Viewer
Open Full PDF

BUILDER'S SANDBOX

Build This Paper

Use an AI coding agent to implement this research.

OpenAI Codex
OpenAI CodexAI Agent

Lightweight coding agent in your terminal.

Claude Code
Claude CodeAI Agent

Agentic coding tool for terminal workflows.

AntiGravity IDE
AntiGravity IDEScaffolding

AI agent mindset installer and workflow scaffolder.

Cursor
CursorIDE

AI-first code editor built on VS Code.

VS Code
VS CodeIDE

Free, open-source editor by Microsoft.

MVP Investment

$9K - $12K
6-10 weeks
Engineering
$8,000
Cloud Hosting
$240
SaaS Stack
$300
Domain & Legal
$100

6mo ROI

2-4x

3yr ROI

10-20x

Lightweight AI tools can reach profitability quickly. At $500/mo average contract, 20 customers = $10K MRR by 6mo, 200+ by 3yr.

Talent Scout

Y

Yinghong Yu

ELLIS Institute Finland, Aalto University

G

Guangyuan Li

Aalto University

J

Jiancheng Yang

ELLIS Institute Finland, Aalto University

Find Similar Experts

AI/ML experts on LinkedIn & GitHub

References (34)

[1]
MedNeXt-v2: Scaling 3D ConvNeXts for Large-Scale Supervised Representation Learning in Medical Image Segmentation
2025Saikat Roy, Yannick Kirchhoff et al.
[2]
Revisiting 2D Foundation Models for Scalable 3D Medical Image Classification
2025Han Liu, Bogdan Georgescu et al.
[3]
MedDINOv3: How to adapt vision foundation models for medical image segmentation?
2025Yuheng Li, Yizhou Wu et al.
[4]
DINOv3
2025Oriane Sim'eoni, Huy V. Vo et al.
[5]
Primus: Enforcing Attention Usage for 3D Medical Image Segmentation
2025Tassilo Wald, Saikat Roy et al.
[6]
VideoRoPE: What Makes for Good Video Rotary Position Embedding?
2025Xilin Wei, Xiaoran Liu et al.
[7]
Does DINOv3 Set a New Medical Vision Standard?
2025Che Liu, Yinda Chen et al.
[8]
Large-Scale 3D Medical Image Pre-Training With Geometric Context Priors
2024Linshan Wu, Jiaxin Zhuang et al.
[9]
Multi-task learning for medical foundation models
2024Jiancheng Yang
[10]
Knowledge transfer from macro-world to micro-world: enhancing 3D Cryo-ET classification through fine-tuning video-based deep models
2024Sabhay Jain, Xingjian Li et al.
[11]
nnU-Net Revisited: A Call for Rigorous Validation in 3D Medical Image Segmentation
2024Fabian Isensee, Tassilo Wald et al.
[12]
Generalist foundation models from a multimodal dataset for 3D computed tomography.
2024I. Hamamci, Sezgin Er et al.
[13]
InternVideo2: Scaling Foundation Models for Multimodal Video Understanding
2024Yi Wang, Kunchang Li et al.
[14]
Medical SAM Adapter: Adapting Segment Anything Model for Medical Image Segmentation
2023Junde Wu, Rao Fu et al.
[15]
Segment Anything
2023A. Kirillov, Eric Mintun et al.
[16]
Segment Anything in Medical Images
2023Jun Ma, Bo Wang
[17]
Masked Autoencoders Are Scalable Vision Learners
2021Kaiming He, Xinlei Chen et al.
[18]
MedMNIST v2 - A large-scale lightweight benchmark for 2D and 3D biomedical image classification
2021Jiancheng Yang, Rui Shi et al.
[19]
Asymmetric 3D Context Fusion for Universal Lesion Detection
2021Jiancheng Yang, Yi He et al.
[20]
LoRA: Low-Rank Adaptation of Large Language Models
2021J. Hu, Yelong Shen et al.

Showing 20 of 34 references

Founder's Pitch

"Develop 3D-enabled AI models from existing 2D models without retraining, leveraging PlaneCycle's adapter-free technology."

AI/MLScore: 8View PDF ↗

Commercial Viability Breakdown

0-10 scale

High Potential

2/4 signals

5

Quick Build

4/4 signals

10

Series A Potential

4/4 signals

10

Sources used for this analysis

arXiv Paper

Full-text PDF analysis of the research paper

GitHub Repository

Code availability, stars, and contributor activity

Citation Network

Semantic Scholar citations and co-citation patterns

Community Predictions

Crowd-sourced unicorn probability assessments

Analysis model: GPT-4o · Last scored: 3/4/2026

🔭 Research Neighborhood

Generating constellation...

~3-8 seconds

Why It Matters

This research allows leveraging vast existing investments in 2D foundation models by enabling them to handle 3D data without retraining, thus saving resources and time especially critical in fields like medical imaging.

Product Angle

Develop an API that enables existing AI models to extend their capabilities to 3D datasets, primarily targeting industries with volumetric data requirements such as healthcare and scientific research.

Disruption

It replaces the need for specialized 3D models or adapting 2D models with additional computational layers, offering a streamlined process for enabling 3D capability.

Product Opportunity

As medical imaging and similar fields increasingly rely on 3D data processing, the ability to retrofit existing 2D models creates a significant market, potentially a multi-billion dollar industry as it affects major imaging device manufacturers and healthcare systems.

Use Case Idea

Adapt medical imaging models to efficiently process CT and MRI scans without needing to develop entirely new 3D models, saving time and computational resources.

Science

PlaneCycle cyclically distributes spatial aggregation across different planes using the original 2D backbone, allowing 3D fusion without architectural modifications or retraining, preserving pretrained 2D model biases.

Method & Eval

The method was evaluated on six 3D classification and three 3D segmentation tasks, showing superior performance over traditional methods both in zero-training and fine-tuning conditions, achieving close to fully trained 3D models.

Caveats

Performance could vary with model architecture; although promising, results depend on the compatibility of existing 2D models with PlaneCycle's method.

Author Intelligence

Yinghong Yu

ELLIS Institute Finland, Aalto University

Guangyuan Li

Aalto University

Jiancheng Yang

ELLIS Institute Finland, Aalto University
jiancheng.yang@aalto.fi