TY - GEN
T1 - Adaptive Conformal Prediction for Reliable and Explainable Medical Image Classification
AU - Octadion, One
AU - Yudistira, Novanto
AU - Muflikhah, Lailil
N1 - Publisher Copyright:
© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2027.
PY - 2027
Y1 - 2027
N2 - Deep learning models for medical imaging often exhibit overconfidence, creating safety risks in ambiguous diagnostic scenarios. While Conformal Prediction (CP) provides distribution-free statistical guarantees, standard methods like Regularized Adaptive Prediction Sets (RAPS) optimize for average efficiency, masking severe failures on difficult inputs. We propose an Adaptive Lambda Criterion for RAPS that minimizes the worst-case coverage violation across prediction set size strata. On OrganAMNIST (58,850 abdominal CT images, 11 classes), standard size-optimized RAPS converges to near-deterministic behavior with stratified undercoverage on uncertain samples, while our method achieves 95.72% global coverage with average set size 1.09 and ≥90% coverage across all strata. Cross-domain validation on PathMNIST (107,180 pathology images, 9 classes) confirms generalizability. Quantitative Grad-CAM analysis (ρ=-0.30, p<10-22) confirms that multi-label predictions correspond to focused attention on anatomically ambiguous regions.
AB - Deep learning models for medical imaging often exhibit overconfidence, creating safety risks in ambiguous diagnostic scenarios. While Conformal Prediction (CP) provides distribution-free statistical guarantees, standard methods like Regularized Adaptive Prediction Sets (RAPS) optimize for average efficiency, masking severe failures on difficult inputs. We propose an Adaptive Lambda Criterion for RAPS that minimizes the worst-case coverage violation across prediction set size strata. On OrganAMNIST (58,850 abdominal CT images, 11 classes), standard size-optimized RAPS converges to near-deterministic behavior with stratified undercoverage on uncertain samples, while our method achieves 95.72% global coverage with average set size 1.09 and ≥90% coverage across all strata. Cross-domain validation on PathMNIST (107,180 pathology images, 9 classes) confirms generalizability. Quantitative Grad-CAM analysis (ρ=-0.30, p<10-22) confirms that multi-label predictions correspond to focused attention on anatomically ambiguous regions.
KW - Conformal Prediction
KW - Explainable AI
KW - Medical Image Classification
KW - Stratified Coverage
KW - Uncertainty Quantification
UR - https://www.scopus.com/pages/publications/105046181520
U2 - 10.1007/978-981-92-2891-1_6
DO - 10.1007/978-981-92-2891-1_6
M3 - Conference contribution
AN - SCOPUS:105046181520
SN - 9789819228904
T3 - Lecture Notes in Computer Science
SP - 70
EP - 81
BT - Advances and Trends in Artificial Intelligence. Theory and Applications - 39th International Conference on Industrial, Engineering and Other Applications of Applied Intelligent Systems, IEA/AIE 2026, Proceedings
A2 - Fujita, Hamido
A2 - Selamat, Ali
A2 - Ghazali, Masitah
A2 - Ali, Moonis
PB - Springer Science and Business Media Deutschland GmbH
T2 - 39th International Conference on Industrial, Engineering and Other Applications of Applied Intelligent Systems, IEA/AIE 2026
Y2 - 6 July 2026 through 8 July 2026
ER -