search

suinleelab / MONET

Transparent medical image AI via an image–text foundation model grounded in medical literature
Other
49 stars 4 forks source link

Access for downsteam-task data meta file #9

Open XiejiLi opened 2 months ago

XiejiLi commented 2 months ago

Hi, authors, Thank you for you work for explainable AI in dermoatology filed. Currently I am reproducing your result in multiple downstream tasks. I tried to dive into a notebook, but it's hard to follow your work without a downstream dataset based on F17k and DDI, would you offer the meta file or code to process these downsteam-task dataset? Thank a lot for presenting a inspired work for us.

XiejiLi commented 2 months ago

Also, I reproduce the result of F17k disease label annotation. When i processed f17k dataset,the positive example is less than 30 being exlcuded, these label left:

melanoma nevus basal cell carcinoma actinic keratosis benign keratosis dermatofibroma vascular lesion malignant seborrheic keratosis squamous cell carcinoma lentigo NOS lichenoid keratosis

there are still 2 labels not found(currently 12 labels in 46051 images), could you provide the label columns you use in this experiment?

chanwkimlab commented 2 months ago

Hi, thank you for your interest in our work! The Fitzpatrick17k dataset contains three columns related to skin conditions: label, nine_partition_label, and three_partition_label. We used the label column, as it provides the most fine-grained disease classification. Also, you can find scripts for preprocessing datasets here. I hope this helps!