PhenoCluster

1. Core Technology

• Intelligent Biomarker Prediction Engine

• Pre-trained models directly process raw images to predict marker positivity probability (0-1), no manual thresholding required.

• Noise-robust training minimizes signal spillover, background noise, and low-contrast impact.

• Semi-automated Cell Classifier

• Infers cell types (e.g., tumor/immune cells) via probability matrices and custom logic tables.

• Compatible with cross-platform data (optical microscopy/MIBI-TOF).

2. Competitive Edge

3. Key Performance

• Speed: 10× faster per image (minutes for end-to-end analysis).

• Accuracy: F1-score 0.75-0.8 (matches expert standards).

• Rare Cell Detection: Identifies novel subtypes via unsupervised processing.

4. Use Cases

• Cancer Research: Deciphering immune infiltration and proliferation markers in TME.

• Immunology: High-resolution T-cell subset profiling (CD4+/CD8+/Treg).

Ⅰ. Analysis of Core Technologies

1. Intelligent Biomarker Prediction Engine

• Pre - trained Model: It can directly input raw images without manual operation and predict the positive probability (ranging from 0 to 1) of biomarkers in any channel, simplifying the pre - analysis preparation and reducing labor and time costs.

• Anti - interference Design: Through noise - robust training, it effectively suppresses the interference of signal overflow, background noise, and low contrast. Even in complex imaging environments, it can stably output reliable biomarker probabilities and ensure the quality of basic analysis data.

• Cross - platform Compatibility: It fully supports multiple imaging platforms such as optical microscopes and mass cytometry by time - of - flight (MIBI - TOF), adapts to different experimental scenarios and data sources, and improves technology reusability.

2. Semi - automated Cell Type Classifier

• Self - Organizing Map (SOM): It uses an unsupervised clustering method to reduce the dimensionality of high - dimensional data and visualize it, helping users intuitively understand the distribution characteristics of cell populations and laying a foundation for subsequent precise classification.

• Dynamic Scoring Function: Based on prior knowledge (such as logic tables), it assigns probability scores to various cell types, abandoning the traditional hard threshold limitation, adapting to complex and diverse cell phenotypes, and improving classification flexibility and accuracy.

• Interactive Exploration: It is deeply integrated with Qupath, supporting real - time result visualization, dynamic adjustment of probability thresholds, and quality control. Researchers can optimize the classification strategy iteratively while analyzing, achieving efficient iteration.

3. Workflow Logic: First, the intelligent biomarker prediction engine generates a single - cell biomarker positive probability matrix, and then the classifier combines the probability matrix with the logic table to automatically infer cell types such as tumor cells and immune cells. The process is clear and closely connected, realizing the efficient transformation from image input to cell type output.

Ⅱ. Core Advantages Compared with Traditional Solutions (Taking Qupath as an Example)