arXiv cs.AI· S. M. Mahbub Uddin Ahmed·· 3 天前AI 评分22
Clef-Flash 决策基础模型如何实现 24 类疾病分类与症状级反事实解释
Toward System One Clinical Intelligence: Parameter-Efficient Fine-Tuning and Counterfactual Explainability of Decision Foundation Models
AI 导读
基于 Qwen3.5 的 90 亿参数决策模型 Clef-Flash 经 4-bit QLoRA 微调后，在 Symptom2Disease 基准的 24 类疾病分类上达到 95.42% 诊断准确率和 0.9551 的 macro F1，超过 ClinicalBERT 的 90.90% 及零样本基线 61.67% 和 65.00%。
正文 · 原文
Abstract:Automated disease classification from patient-reported symptoms requires balancing the speed of lightweight discriminative models against the contextual understanding and latency costs of autoregressive large language models (LLMs). Decision foundation models offer an alternative by mapping representations directly to candidate classes without free-form text generation. We adapt Clef-Flash, a 9-billion-parameter decision model built on a Qwen3.5 backbone, for 24-class disease classification on the Symptom2Disease benchmark. Using 4-bit quantized low-rank adaptation (QLoRA) with differential learning rates and a stratified split of 960 training and 240 test cases, the model achieves 95.42% diagnostic accuracy and a macro F1 score of 0.9551. This improves upon the reported zero-shot baselines of 61.67% for a remote API and 65.00% for a local unadapted initialization, and outperforms the ClinicalBERT baseline (90.90%). By evaluating candidate logits in a single forward pass without autoregressive decoding, inference takes under 80 milliseconds per record on a single GPU, yielding a reported 31-fold speedup over generative LLM baselines. To examine symptom-level evidence, we introduce a Leave-One-Symptom-Out (LOSO) counterfactual perturbation framework. Removing disease-defining symptom clauses in a psoriasis example reduces predicted-class confidence from 100.00% to 0.12% and shifts the prediction to impetigo with 99.59% confidence, while removing secondary systemic symptoms produces less than 0.01% change in confidence. These findings suggest that decision-based architectures can combine the linguistic capabilities of large foundation models with low-latency inference and symptom-level interpretability for clinical triage.
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.15029 [cs.CL] |
| (or arXiv:2610.15029v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.15029 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: S. M. Mahbub Uddin Ahmed [view email]
[v1]
Sat, 10 Oct 2026 16:13:49 UTC (176 KB)
来源：arXiv cs.AI · arxiv.org