A year ago, I developed a study prototype of a neural network that combines two types of data:
- medical images in DICOM format;
- clinical tabular data (patient age, tumor size, biopsy results).
The goal of the model is to analyze both images and numerical data simultaneously to classify cancer presence.
Key Features
- Multimodality: the model processes both images and tabular features.
- Attention mechanism: highlights the most important features to improve accuracy.
- GPU/CPU support: training can be performed on a regular computer or on a GPU.
- Evaluation metrics: AUC, F1, Precision, Recall — to measure performance objectively.
- Engineering design: separate classes for dataset, model, training, and logging.
In summary: this project gave me hands‑on experience with medical data and showed how Python can be applied not only in backend development but also in machine learning tasks.
Top comments (0)