A 10-year-old CNN outperforms frontier LLMs for skin cancer diagnosis
A 10-year-old CNN outperforms frontier LLMs for skin cancer diagnosis
我向 GPT-6 Astra 和 Claude Fable 5.1（2026 年 10 月市面上最强大的两个 AI 模型）展示了两张图片：一张临床照片和一张皮肤镜（皮肤显微镜）图像。两者都认为这是脂溢性角化病，一种无害的疣状增生，并表示患者可以出院。随后，我将这些图片展示给经过微调的 Inception v3 版本，这是一个由谷歌于 2015 年首次发布的卷积神经网络（CNN），并在此处使用公开的皮肤病图像进行了微调。该 CNN 根据皮肤镜视图得出了正确答案。
I showed these two images - a clinical photograph and a dermoscopic (skin microscope) image - to GPT-6 Astra and Claude Fable 5.1, two of the most capable AI models available in October 2026. Both said it was a seborrhoeic keratosis, a harmless warty growth, and that the patient could be discharged. I then showed them to fine-tuned version of Inception v3, a convolutional neural network (CNN) first published by Google in 2015 and fine-tuned here on public dermatology images. The CNN got the right answer based on the dermoscopy view.
这是我提供给这三个模型的 200 个经活检证实的病变之一。CNN 对其中 125 个（62%）给出了正确诊断（在 6 种可能性中）。两个前沿 LLM 分别给出了 84 个和 85 个（42%）正确诊断。一个经过约二十分钟微调的十年历史的网络，在识别皮肤癌方面击败了它们两者。
It was one of 200 biopsy-proven lesions I gave to all three. The CNN named the right diagnosis (out of 6 possiblities) for 125 of them (62%). The two frontier LLMs named it for 84 and 85 (42%). A ten-year-old network, fine-tuned for about twenty minutes, beat both of them at recognising skin cancer.
这些病变来自 MILK10k，这是一个 2025 年通过国际皮肤成像协作组织发布的公开数据集。其中每个病变都被拍摄了两次：一张临床特写和一张通过皮肤科医生使用的照明放大镜拍摄的皮肤镜图像。我随机抽取了 200 个病变，所有病变均经组织病理学证实：
The lesions come from MILK10k, a public dataset released through the International Skin Imaging Collaboration in 2025, in which each lesion is photographed twice: a clinical close-up and a dermoscopic image taken through the illuminated magnifying lens dermatologists use. I drew 200 lesions at random, all confirmed by histopathology:
每个病变都进行了活检，包括良性病变，因此这些病变更难诊断：有人决定切除的痣是临床医生感到担忧的病变。
Every lesion was biopsied, including the benign ones, so these are harder to diagnose: a mole that somebody decided to cut out is one that worried a clinician.
一名 75 岁男性的躯干基底细胞癌。所有三个模型都回答正确。
Basal cell carcinoma on the trunk of a 75-year-old man. All three models answered correctly.
厚度小于 0.8 毫米的浸润性黑色素瘤，色素很少。GPT-6 Astra 以 95% 的置信度将其称为基底细胞癌。Claude Fable 5.1 和 CNN 将其称为黑色素瘤。
Invasive melanoma under 0.8 mm thick, with very little pigment. GPT-6 Astra called it a basal cell carcinoma with 95% confidence. Claude Fable 5.1 and the CNN called it a melanoma.
腿部的脂溢性角化病。两个 LLM 都将其称为光化性角化病，而 CNN 将其称为鳞状细胞癌：三者都错了，尽管三者都会建议保留它而不是让其出院。
Seborrhoeic keratosis on the leg. Both LLMs called it an actinic keratosis and the CNN a squamous cell carcinoma: all three were wrong, though all three would have kept it rather than discharge it.
LLM 仅被提供了两张图像，并被要求像临床医生使用它们那样，从六种诊断中选出最可能的一种（或“不确定”），给出置信度、第二选择，以及该病变是否可以在无需治疗、活检或随访的情况下安全出院。两个模型在默认设置下收到了相同的提示，且没有拒绝任何病例。
The LLMs were given both images and nothing else, and asked, as a clinician might use them, for the most likely of the six diagnoses (or “unsure”), a confidence, a second choice, and whether the lesion could be safely discharged with no treatment, biopsy or follow-up. Both models got the same prompt at default settings, and neither refused a case.
该 CNN 特意没有采用最先进的技术。我使用了在日常照片上预训练的 Inception v3，然后在单块 H100 GPU 上使用来自 ISIC 档案的 68,011 张皮肤镜和临床图像对所有层进行了微调。这 200 个测试病变以及同一病变或患者的所有其他照片在训练前都已从其数据（包括验证集）中删除，因此它从未见过它们，且它们未被用于选择模型或设置其阈值。它对每个病变的这两张图像进行评分，将它们结合起来（皮肤镜图像权重较大），并从相同的六种诊断中进行选择。
The CNN is deliberately not state of the art. I used Inception v3 pretrained on everyday photographs and then fine-tuned all layers on 68,011 dermoscopic and clinical images from the ISIC Archive on a single H100 GPU. These 200 test lesions, and every other photograph of the same lesion or patient, were removed from its data (including the validation set) before training, so it had never seen them, and they were not used to choose the model or set its threshold. It scored both images of each lesion, combined them with most of the weight on the dermoscopic image, and chose from the same six diagnoses.
你可以亲自尝试这个 CNN。值得注意的是，该模型足够小，可以在手机浏览器中使用 WebGPU 完全运行。没有任何内容被上传，其随机样本按钮从这 200 个病变中进行抽取。
You can try the CNN yourself. Remarkably the model is small enough to run entirely in a phone browser using WebGPU. Nothing is uploaded, and its random sample button draws from these same 200 lesions.
总体而言，CNN 在大多数诊断中表现更准确。允许第二选择时，它达到了 200 个中的 173 个，而另外两个模型分别为 134 和 138 个。在 30 个黑色素瘤中，它识别出了 13 个，而另外两个分别为 8 个和 11 个。最大的差距在于无害病变：它识别出了 50 个痣中的 37 个和 50 个脂溢性角化病中的 26 个，而 GPT-6 Astra 和 Claude Fable 5.1 分别为 16 和 13 个，以及 33 和 6 个。在基底细胞癌方面，它比 GPT-6 Astra 少一个，在光化性角化病方面落后于两个 LLM。
The CNN was right more often overall and on most diagnoses. Allowing a second choice, it reached 173 of 200 against 134 and 138. It named 13 of the 30 melanomas against 8 and 11. The biggest gap was on the harmless lesions: it recognised 37 of 50 moles and 26 of 50 seborrheic keratoses, against 16 and 13 for GPT-6 Astra and 33 and 6 for Claude Fable 5.1. It was one short of GPT-6 Astra on basal cell carcinoma and behind both LLMs on actinic keratosis.
这些模型的失败方式也不同。每个 LLM 都有一个偏好的答案：GPT-6 Astra 将 75 个病变称为基底细胞癌（实际只有 25 个），其中包括 30 个黑色素瘤中的 14 个；Claude Fable 5.1 将 53 个病变称为光化性角化病（实际只有 20 个）。CNN 在黑色素瘤上的错误则相反：在它判断错误的 17 个病例中，它将 8 个称为痣，6 个称为脂溢性角化病。
The models also failed differently. Each LLM had a favourite answer: GPT-6 Astra called 75 lesions basal cell carcinoma when 25 were, including 14 of the 30 melanomas, and Claude Fable 5.1 called 53 lesions actinic keratosis when 20 were. The CNN’s errors on melanoma went the other way: of the 17 it got wrong, it called 8 a naevus and 6 a seborrhoeic keratosis.
如果以患者身份询问，仅提供临床照片并提问“我需要看医生吗？”，两个 LLM 在几乎所有情况下都建议看医生，包括无害病变：在另外 100 个经活检证实的黑色素瘤中，它们分别建议了 99 和 98 次。
Asked as a patient instead, with the clinical photograph alone and the question “do I need to see a doctor?”, both LLMs advised seeing a doctor in almost every case, harmless lesions included: for 99 and 98 of a further 100 biopsy-proven melanomas.
一个使用过时架构的小型、老旧、专业化网络在识别皮肤病变方面远胜于两个前沿 LLM。
A small, old, specialised network using an outdated architecture recognises skin lesions far better than two frontier LLMs.
病例、提示词、所有 LLM 回复和评分代码都在一个下载包中，CNN 在这里。
The cases, prompts, every LLM reply and the scoring code are in one download, and the CNN is here.
我设计了这个测试并做出了临床裁定。分析和初稿是使用 Anthropic 的编码工具 Claude Code 制作的，该工具运行在 Claude 模型上；测试的两个 LLM 之一也是 Anthropic 的产品。评分是通过针对组织学结果的脚本完成的。
I designed the test and made the clinical rulings. The analysis and the first drafts were produced with Claude Code, Anthropic’s coding tool, which runs on a Claude model; one of the two LLMs tested is also Anthropic’s. Scoring was done by script against the histology.
图片来自 MILK10k 数据集（MILK 研究团队，国际皮肤成像协作组织），根据 CC BY-NC 许可使用。
Images are from the MILK10k dataset (MILK study team, International Skin Imaging Collaboration), used under a CC BY-NC licence.
Magnus Lynch 博士是盖伊医院的皮肤科顾问医生和莫氏外科医生，也是伦敦国王学院的荣誉讲师（相当于副教授）。他的临床工作涵盖皮肤癌、莫氏手术、面部重建手术和激光；他的研究涵盖皮肤癌和疤痕的细胞生物学以及人工智能在皮肤病学中的应用。完整资料。
Dr Magnus Lynch is a Consultant Dermatologist and Mohs Surgeon at Guy’s Hospital and Honorary Reader (equivalent of Associate Professor) at King’s College London. His clinical work covers skin cancer, Mohs and facial reconstructive surgery, and lasers; his research covers the cell biology of skin cancers and scars and the use of AI in dermatology. Full profile.