1. Choose the diagnostic condition

The NIH ChestX-ray14 dataset comprises over 100,000 anonymized, frontal-view chest radiographs. The dataset includes labels for distinct conditions, which were derived from the original radiology reports using Natural Language Processing (NLP) techniques. The 40 loaded images are tagged as

present

or

absent

based on the reference labels for this condition.

Compute Embeddings

2. Make adjustments to the training set

2. Change text prompts

Images marked with Training or Evaluation can be adjusted as a percentage

50%

Train and Run Classifier

Type or select the diagnostic condition Select a positive text prompt. "Positive" means that the condition is present.

⋁

Select a negative text prompt. "Negative" means that the condition isn't present.

⋁

Evaluate

3. Review performance and make adjustments to the model threshold

Adjusting the threshold affects the balance between false positives and false negatives

50%