MalariAI精准分割疟原虫细胞并解释分类结果，适合临床审计。
MalariAI: A Label-Resilient Decoupled Framework for Annotation-Agnostic Cell Segmentation and Explainable Stage Classification in Dense Malaria Blood Smears
- 分两阶段解耦处理：先无标注分割细胞，再分类
- 细胞分割召回率达75.95%，分类准确率98.36%
- 生成热力图支持个体级诊断验证，适合医疗审核
从血涂片显微镜图像自动诊断疟疾是全球健康领域的重要挑战，专家短缺仍是主要瓶颈。现有深度学习系统存在三大累积性缺陷：端到端检测器将未标注细胞误判为背景，导致召回率受标注完整性影响而非真实细胞恢复；非极大值抑制在密集血涂片中压制有效检测；且流程缺乏每细胞的时空证据以供临床审计。本文提出MalariAI，一种两阶段解耦框架，解决上述问题。第一阶段采用无标注水文算法，在完整1600x1200图像上分离所有细胞，无需任何真实标注即可恢复75.95%的真实细胞。整体管道在二分类寄生虫[email protected]上达到29.10%（临床关键指标），多类别[email protected]为8.67%，主要因水文算法生成的有机区域边界与轴对齐标注框不匹配，并非定位失败。第二阶段使用带焦点损失的EfficientNet-B0在真实标注图像块上微调，实现98.36%分类准确率——一旦细胞被正确定位，即为理论上限。稀有阶段（裂殖体、配子体）准确率分别为87.5%和75.0%，远超同一类别的YOLOv8s检测器（分别38.45%和57.27% AP）。每个检测细胞生成的Grad-CAM++热图提供实例级空间证据，定量分析显示激活能量显著高于几何随机基线（+0.0485，配对p=1.4×10⁻³³），使显微镜医师可在不牺牲性能的前提下逐个验证预测结果。
原文摘要 · Abstract (English)
Automated malaria diagnosis from blood smear microscopy is a critical global health AI challenge; expert scarcity remains the primary diagnostic bottleneck. Existing deep learning systems face three compounding failures: end-to-end detectors treat unannotated cells as background, skewing recall by annotation completeness rather than true cell recovery; Non-Maximum Suppression suppresses valid detections in dense smears; and pipelines lack per-cell spatial evidence for clinical audit. We present MalariAI, a two-stage decoupled framework addressing all three. Stage 1 applies an annotation-agnostic watershed algorithm to isolate every cell in a full 1600x1200 image, recovering 75.95% of ground-truth cells without any ground-truth input. End-to-end, the pipeline reaches a binary parasitized [email protected] of 29.10% - the clinically relevant metric for flagging any infected cell - while the stricter multi-class [email protected] of 8.67% mainly reflects watershed's organic region boundaries being penalized against axis-aligned ground-truth boxes, not a localisation failure. Stage 2 fine-tunes EfficientNet-B0 with Focal Loss on ground-truth crops, achieving 98.36% classification accuracy - an oracle upper bound once a cell is correctly localised - with 87.5% and 75.0% accuracy on the rare schizont and gametocyte stages, versus 38.45% and 57.27% AP for a modern YOLOv8s detector evaluated end-to-end on the same classes. Grad-CAM++ heatmaps generated per detected cell provide instance-level spatial evidence for clinical audit; a quantitative energy-in-box analysis confirms this activation is concentrated on the annotated cell body significantly above a geometric chance baseline (+0.0485, paired p = 1.4 x 10^-33), letting microscopists verify predictions at the individual parasite level without sacrificing classification performance.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。