MorphXAI: An Explainable Framework for Morphological Analysis of Parasites in Blood Smear Images

TL;DR

The paper tackles the need for interpretable automated parasite detection in blood smears by integrating morphological reasoning into the detection pipeline. It introduces MorphXAI, a morphology-guided extension of RT-DETRv3 that adds a Morphological Decoder to predict structured attributes (shape, curvature, dot count, flagellum, developmental stage) alongside bounding boxes; this is supported by a clinician-annotated dataset for three parasite species. A GT2MorphologyTarget data pipeline and a joint optimization ensure explanations are faithful to expert reasoning while preserving detection accuracy. Empirically, MorphXAI achieves a modest gain in $AP^{.50:.95}$ over a strong baseline and yields high attribute accuracies with real-time performance, demonstrating that clinically meaningful explanations can be embedded into automated parasite analysis.

Abstract

Parasitic infections remain a pressing global health challenge, particularly in low-resource settings where diagnosis still depends on labor-intensive manual inspection of blood smears and the availability of expert domain knowledge. While deep learning models have shown strong performance in automating parasite detection, their clinical usefulness is constrained by limited interpretability. Existing explainability methods are largely restricted to visual heatmaps or attention maps, which highlight regions of interest but fail to capture the morphological traits that clinicians rely on for diagnosis. In this work, we present MorphXAI, an explainable framework that unifies parasite detection with fine-grained morphological analysis. MorphXAI integrates morphological supervision directly into the prediction pipeline, enabling the model to localize parasites while simultaneously characterizing clinically relevant attributes such as shape, curvature, visible dot count, flagellum presence, and developmental stage. To support this task, we curate a clinician-annotated dataset of three parasite species (Leishmania, Trypanosoma brucei, and Trypanosoma cruzi) with detailed morphological labels, establishing a new benchmark for interpretable parasite analysis. Experimental results show that MorphXAI not only improves detection performance over the baseline but also provides structured, biologically meaningful explanations.

Figures (6)

• Figure 1: (a) Existing approaches rely on post-hoc XAI, producing heatmaps without clinical interpretability (gap). (b) Our proposed framework introduces an XAI method that directly outputs structured morphological attributes, interpretable by clinicians.
• Figure 2: Overview of MorphXAI. The framework introduces a morphological decoder that jointly predicts bounding boxes, parasite class, and clinically relevant attributes. A dedicated data pipeline converts clinician annotations into supervision targets, and a morphology-aware optimization guides training. The model produces both detections and structured morphological reports, enabling interpretable parasite analysis.
• Figure 3: Inference pipeline of MorphXAI. Given a microscopic blood smear image, our framework detects parasites and augments each prediction with structured, clinically meaningful explanations. The output includes parasite type, morphological traits, and confidence scores, providing interpretable results aligned with clinical reasoning.
• Figure 4: Clinician-verified morphological attributes used in our dataset. These structured attributes capture the key cues that form the basis for our explainability framework.
• Figure 5: Morphological attribute, Curvature, was defined strictly on the parasite body. Clinician annotators labeled the curvature attribute based only on the parasite body, without considering the orientation of the free-floating flagellum. In images (b) and (c), the body is clearly C-shaped, but in (b) the flagellum extends outward in an S-like curve. This makes it visually similar to (a), where the parasite body itself is S-shaped. Such cases illustrate the biological complexity of curvature assessment. These visual overlaps explain why curvature prediction accuracy is relatively low. With larger datasets and refined attribute definitions, this challenge can be further overcome.