A convolutional network Google published in 2015 diagnosed skin lesions correctly 62% of the time. GPT-6 Astra and Claude Fable 5.1 each scored 42%.
A convolutional network Google published in 2015 diagnosed skin lesions correctly 62% of the time. GPT-6 Astra and Claude Fable 5.1 each scored 42%.
0 vue
Publié le 26/10/2026 14:21
1:32
A convolutional network Google published in 2015 diagnosed skin lesions correctly 62% of the time. GPT-6 Astra and Claude Fable 5.1 each scored 42%.
1 commentaire
?
Dermatologist Magnus Lynch tested the two LLMs and the CNN, Inception v3, on 200 biopsy-proven lesions from the MILK10k dataset, each with a clinical photo and dermoscopic image. Same prompts, default settings.
Lynch fine-tuned Inception v3 on 68,011 ISIC Archive images, about 20 minutes on one H100 GPU. The 200 test lesions were excluded from that training set.