Robust Fair Disease Diagnosis in CT Images

Justin Li; Daniel Ding; Asmita Yuki Pritha; Aryana Hou; Xin Wang; Shu Hu

cs.CVApr 8, 2026

Robust Fair Disease Diagnosis in CT Images

Authors: Justin Li, Daniel Ding, Asmita Yuki Pritha, Aryana Hou, Xin Wang, Shu Hu

Organizations: Carmel High School, Carmel, Indiana, USA · Capstone School Dhaka, Bangladesh · Clarkstown High School South, West Nyack, New York, USA · University at Albany, State University of New York, Albany, New York, USA · Purdue University, West Lafayette, Indiana, USA

Abstract

Automated diagnosis from chest CT has improved considerably with deep learning, but models trained on skewed datasets tend to perform unevenly across patient demographics. However, the situation is worse than simple demographic bias. In clinical data, class imbalance and group underrepresentation often coincide, creating compound failure modes that neither standard rebalancing nor fairness corrections can fix alone. We introduce a two-level objective that targets both axes of this problem. Logit-adjusted cross-entropy loss operates at the sample level, shifting decision margins by class frequency with provable consistency guarantees. Conditional Value at Risk aggregation operates at the group level, directing optimization pressure toward whichever demographic group currently has the higher loss. We evaluate on the Fair Disease Diagnosis benchmark using a 3D ResNet-18 pretrained on Kinetics-400, classifying CT volumes into Adenocarcinoma, Squamous Cell Carcinoma, COVID-19, and Normal groups with patient sex annotations. The training set illustrates the compound problem concretely: squamous cell carcinoma has 84 samples total, 5 of them female. The combined loss reaches a gender-averaged macro F1 of 0.8403 with a fairness gap of 0.0239, a 13.3% improvement in score and 78% reduction in demographic disparity over the baseline. Ablations show that each component alone falls short. The code is publicly available at https://github.com/Purdue-M2/Fair-Disease-Diagnosis.

Figures & tables

Figure 1: Training set distribution by disease category and patient sex. The dataset exhibits a compound imbalance: squamous cell carcinoma is already the rarest class overall (84 samples), and its female subset contains only 5 training examples. This intersection of class rarity and demographic scarcity motivates our two-level objective that simultaneously addresses both axes of imbalance.

Figure 2: Overview of the proposed framework. Row 1 : representative CT slices from the four diagnostic categories. Row 2 : preprocessing removes non-pulmonary regions, samples to a fixed depth of 64, and normalizes to produce the input volume. Row 3 : a Kinetics-400 pretrained R3D-18 extracts features, classified through a fully connected head. The logit adjusted loss (§ 3.4) corrects class imbalance, while CVaR aggregation (§ 3.5) enforces gender fairness. Dashed arrows indicate metadata inputs.

Figure 3: Performance vs. fairness trade-off across all configurations. Circles ( $\bullet$ ) denote FairLACVaR at different $\alpha$ values, diamonds denote single-component ablations, and the square marks the baseline. The star ( $\star$ ) indicates the best overall configuration ( $\alpha=0.8$ ), which occupies the upper-left region corresponding to high score and low fairness gap.

Baseline & Ablations
Method	$\alpha$	F1 male	F1 female	Score $\uparrow$	Gap $\downarrow$
Baseline (CE)	–	0.7957	0.6868	0.7413	0.1089
LA Only	–	0.8596	0.7375	0.7986	0.1221
CVaR Only	0.7	0.8738	0.7591	0.8165	0.1148
FairLACVaR (LA + CVaR)
LA + CVaR	0.4	0.7574	0.7185	0.7379	0.0389

Table 1: Main results and ablation study on the challenge validation set. All runs use a fixed random seed on NVIDIA A100. Best values in bold .

$\alpha$	GPU	F1 male	F1 female	Score $\uparrow$	Gap $\downarrow$
0.7	A100	0.7835	0.8394	0.8115	0.0559
0.7	L4	0.8467	0.7415	0.7941	0.1052
0.8	A100	0.8283	0.8522	0.8403	0.0239
0.8	L4	0.7498	0.7591	0.7544	0.0093
0.9	A100	0.8910	0.7611	0.8260	0.1299
0.9	L4	0.8379	0.7813	0.8096	0.0565

Table 2: Hardware sensitivity analysis. Same seed and code on NVIDIA A100 vs. L4. The optimal

\alpha

shifts across platforms, but all configurations within

\alpha\in[0.7,0.9]

improve over the baseline.

Explore similar work

Fair Lung Disease Diagnosis from Chest CT via Gender-Adversarial Attention Multiple Instance Learning
Mar 13, 2026Aditya Parikh, Aasa FeragenMedical Image Classification Algorithmic Fairness
Technical University of Denmark (DTU), Kongens Lyngby, Denmark
Towards Fair and Robust Volumetric CT Classification via KL-Regularised Group Distributionally Robust Optimisation
Mar 16, 2026Samuel Johnny, Blessed Guda, Goodness Obasi +2Medical Image Classification Distributionally Robust Optimization
Carnegie Mellon University Africa, Kigali, Rwanda · Carnegie Mellon University, Pittsburgh, USA
Representation Disentanglement for Fair Chest X-Ray Diagnosis
Oct 5, 2026Yujie Sun, Ruizhe Li, Xiaowu SunChest X-Ray Classification Representation Disentanglement
Academy of Artificial Intelligence and Advanced Technology, Xi’an Jiaotong-Liverpool University, China · Nottingham Biomedical Research Centre, School of Medicine, University of Nottingham, UK

Mar 13, 2026J/K move · Enter open · S save

Fair Lung Disease Diagnosis from Chest CT via Gender-Adversarial Attention Multiple Instance Learning

Aditya Parikh, Aasa Feragen

Technical University of Denmark (DTU), Kongens Lyngby, Denmark

Medical Image Classification Algorithmic Fairness

Open paper AskPDF arXiv

We present a fairness-aware framework for multi-class lung disease diagnosis from chest CT volumes, developed for the Fair Disease Diagnosis Challenge at the PHAROS-AIF-MIH Workshop (CVPR 2026). The challenge requires classifying CT scans into four categories -- Healthy, COVID-19, Adenocarcinoma, and Squamous Cell Carcinoma -- with performance measured as the average of per-gender macro F1 scores, explicitly penalizing gender-inequitable predictions. Our approach addresses two core difficulties: the sparse pathological signal across hundreds of slices, and a severe demographic imbalance compounded across disease class and gender. We propose an attention-based Multiple Instance Learning (MIL) model on a ConvNeXt backbone that learns to identify diagnostically relevant slices without slice-level supervision, augmented with a Gradient Reversal Layer (GRL) that adversarially suppresses gender-predictive structure in the learned scan representation. Training incorporates focal loss with label smoothing, stratified cross-validation over joint (class, gender) strata, and targeted oversampling of the most underrepresented subgroup. At inference, all five-fold checkpoints are ensembled with horizontal-flip test-time augmentation via soft logit voting and out-of-the-fold threshold optimization for robustness. Our model achieves a mean validation competition score of 0.685 (std - 0.030), with the best single fold reaching 0.759. All training and inference code is publicly available at https://github.com/ADE-17/cvpr-fair-chest-ct

Abstract

Explore similar work

Fair Lung Disease Diagnosis from Chest CT via Gender-Adversarial Attention Multiple Instance Learning

Towards Fair and Robust Volumetric CT Classification via KL-Regularised Group Distributionally Robust Optimisation

Representation Disentanglement for Fair Chest X-Ray Diagnosis