This advanced machine learning project tackles a critical challenge in biomedical research: accurately classifying tumor cell lines using microscopy imaging techniques. By leveraging deep learning and advanced image processing, we aim to address the persistent issue of cell line misidentification.
- Develop a robust machine learning model to classify 9 challenging cell lines with high precision
- Demonstrate the power of multi-channel image processing in biological image classification
- Create a generalizable approach for distinguishing between closely related cell lines
| Cell Line | Cancer Origin | Tissue Type |
|---|---|---|
| PC-3 | Prostate Cancer | Epithelial |
| U-251 MG | Glioblastoma | Glial |
| HeLa | Cervical Cancer | Epithelial |
| A549 | Lung Carcinoma | Epithelial |
| U-2 OS | Osteosarcoma | Bone |
| MCF7 | Breast Cancer | Epithelial |
| HEK 293 | Embryonic Kidney | Epithelial |
| CACO-2 | Colorectal Adenocarcinoma | Epithelial |
| RT4 | Bladder Cancer | Epithelial |
Our unique approach combines three grayscale images representing different cellular compartments:
- Nucleus Channel: Structural foundation
- Microtubules Channel: Cellular infrastructure
- Endoplasmic Reticulum Channel: Cellular processing network
- Image Preprocessing: Custom multi-channel image merging
- Data Augmentation: Techniques to enhance model generalizability
- Transfer Learning: Utilizing pre-trained VGG16_bn architecture
- Evaluation Metric: Balanced accuracy for robust performance assessment
- Merge 3 grayscale images into a single RGB representation
- Apply advanced data augmentation
- Normalize image features
- Prepare dataset for deep learning model
- Base Model: VGG16 (Batch Normalization variant)
- Fine-tuned for multi-class cell line classification
- Handles nuanced differences between similar cell lines