● Covered by 1 source · 1 reportMedium impact1 neutral
10-year-old CNN Outperforms Frontier LLMs in Skin Cancer Diagnosis
Key points
- Inception v3 CNN diagnosed 62% of lesions correctly.
- Frontier LLMs diagnosed 42% of lesions correctly.
- CNN was fine-tuned on public dermatology images.
- Test used 200 biopsy-proven skin lesions.
CNN vs. LLM Performance in Diagnosis
A recent comparison showed that a fine-tuned Inception v3, a convolutional neural network (CNN) first published in 2015, outperformed two frontier large language models (LLMs), GPT-6 Astra and Claude Fable 5.1, in diagnosing skin lesions. The CNN correctly identified 125 out of 200 biopsy-proven lesions (62%), whereas the LLMs identified 84 and 85 lesions respectively (42%). This indicates that specialized, older AI architectures can still be more effective for specific tasks than general-purpose, newer models.
Methodology of the Test
The test involved 200 biopsy-proven lesions from the MILK10k public dataset, released in 2025. Each lesion included a clinical photograph and a dermoscopic image. The dataset comprised 100 lesions that required further action (e.g., melanoma, basal cell carcinoma) and 100 benign lesions (e.g., moles, seborrhoeic keratoses). All lesions had been biopsied, including benign ones, suggesting they were clinically concerning cases.
Model Configuration and Training
The LLMs were provided with both images and asked for the most likely diagnosis from six possibilities, a confidence level, a second choice, and whether the lesion could be safely discharged. The Inception v3 CNN was not a state-of-the-art model. It was pretrained on everyday photographs and then fine-tuned on 68,011 dermoscopic and clinical images from the ISIC Archive using a single H100 GPU. The 200 test lesions and any related images were excluded from its training and validation data to ensure an unbiased evaluation.
Implications for Medical AI
The results suggest that while LLMs are versatile, specialized AI models, even older ones, can offer superior performance in specific domains like medical image diagnosis when appropriately fine-tuned. This highlights the importance of task-specific model development and training data in achieving high accuracy for critical applications such as skin cancer detection.
✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →
The daily brief
One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.
One email a day. Unsubscribe in one click, any time.
Today's brief
Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.
▶ Play today's brief
New every morning, and the back catalogue is archived by date.
Reporting from
A fine-tuned Inception v3 convolutional neural network (CNN) from 2015 achieved higher accuracy than current frontier large language models (LLMs) like GPT-6 Astra and Claude Fable 5.1 in diagnosing skin lesions. The CNN correctly identified 62% of 200 biopsy-proven lesions, while the LLMs achieved 42% accuracy, demonstrating the continued relevance of specialized AI for specific medical tasks.