cell-cycle-classification

Environment

• requirements.txt

Directory

• Modify saving directory like checkpoint, record and model via config.py

Step 1: Mitochondrai and nucleus prediction from brightfield

Data

• Set1 Input: brightfield | Label: mitochondria
• Set2 Input: brightfield | Label: nucleus

Method:

• Model: UNet
• Optimizer: Adam
• Loss function: Huber Loss
• Accuracy: Pearson correlation coefficient
• Batch size: 2
• Epoch: 100

Training

• cd train python train_predict_mito_nuclei.py

Result

• UNet Training result
  • Mitochondria
    • Training Pearson Correlation Coefficient: 0.78
    • Validation Pearson Correlation Coefficient: 0.72
  • nucleous
    • Training Pearson Correlation Coefficient: 0.85
    • Validation Pearson Correlation Coefficient: 0.83

Plot

• cd plot python predict_for_mitochondira_nucleus.py python ploy_record.py

Step 2: Cell cycle classification from brightfield/predicted mitochondria/predicted nucleus

Data

• Input: brightfield + predicited mitochondria + predicted nucleus
• Label: 4 class for each pixel (0: background, 1: RFP, 2: GFP, 3: Both)

Method

• Model: Efficient UNet
• Optimizer: Adam
• Loss function: Focal + Dice loss
• Accuracy: F1 score
• Batch size: 4
• Epoch: 100

Training

• cd train python train_cellcycle_phase.py

Result

• Still trying