Skip to content

LexBwmn/ACE-Models

Folders and files

NameName
Last commit message
Last commit date

Latest commit

Β 

History

2 Commits
Β 
Β 

Repository files navigation

# ACE Models: Optimized Brain Tumor Detection

Why I Created ACE

The ACE (Automated Clinical Evaluation) models were developed to solve a critical radiology bottleneck: the widening gap between high-volume imaging demand and the time available for expert human review.

I built these models to prove that high-precision medical computer vision doesn't require massive server infrastructure. My goal was to create efficient diagnostic support that runs locally on a standard MacBook Pro. By prioritizing 1.00 Background Specificity, I designed these models to eliminate the "hallucination" of anomalies in healthy tissue, ensuring reliable performance without the usual computational overhead.


πŸ”— Repository

View these models on Hugging Face: https://huggingface.co/LexBwmn/ACE-V1


πŸ”¬ Model Overview

  • ACE-V1: This is the foundational base model. It is available for download within the archive file located on my Hugging Face profile.
  • ACE-V1.1: This is the current, primary iteration. It features refined architecture and enhanced metrics for optimal performance.

MEDICAL RESEARCH USE ONLY. ACE models are NOT a cleared medical device. They must not be used for primary diagnosis or clinical decision-making. All outputs must be verified by a qualified professional.


πŸ“Š Performance Metrics (ACE-V1.1)

Metric Value
mAP50 0.899
Precision 90.0%
Recall 83.8%
Background Specificity 1.00

βš™οΈ Technical Benchmarks (Peak Performance - V1.1)

  • Mean Pixel Dice (2D): 0.6810
  • Mean Pixel Precision: 0.6645
  • Mean Pixel Recall: 0.7879
  • Inference Speed: 3.95 it/s

πŸ›  Hardware & Environment

  • Training Platform: MacBook Pro (M1 Pro)
  • Acceleration: Apple Silicon Metal Performance Shaders (MPS)
  • Framework: Ultralytics YOLOv11
  • Training Strategy: 120 Total Epochs

πŸ“ Usage Guide

Due to the custom channel architecture required to maintain 1.00 specificity, these models are optimized for the following settings:

  • Confidence Threshold: 0.466
  • Input Image Size: 640x640

Implementation Notice

These models are protected under CC BY-NC-ND 4.0. For institutional adoption, commercial licensing, or clinical integration support, please contact the maintainer.


πŸŽ“ Citation

If you utilize these models in your research, please cite:

@misc{bowman2026ace,
  author = {Bowman, Alexa},
  title = {ACE Models: Optimized Brain Tumor Detection with 1.00 Background Specificity},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{[https://huggingface.co/LexBwmn/ACE-V1](https://huggingface.co/LexBwmn/ACE-V1)}}
}

About

No description, website, or topics provided.

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors