The paper "MambaMIL: Enhancing Long Sequence Modeling with Sequence Reordering in Computational Pathology" introduces a novel approach, MambaMIL, which integrates the Selective Scan Space State Sequential Model (Mamba) into Multiple Instance Learning (MIL) for computational pathology. The primary goal is to address the limitations of existing MIL methods in handling long sequences of instances and mitigating overfitting issues. MambaMIL leverages the capabilities of Mamba to model long sequences efficiently with linear complexity, while introducing the Sequence Reordering Mamba (SR-Mamba) to capture long-range dependencies and improve feature discrimination. SR-Mamba enhances the original Mamba by allowing interactions between instances in both sequential and transpositional orderings, thereby capturing more discriminative features. The method is evaluated on two challenging tasks—survival prediction and cancer subtyping—using nine diverse datasets. Extensive experiments demonstrate that MambaMIL outperforms state-of-the-art MIL methods, showing superior performance in terms of accuracy and stability. The proposed framework is also shown to effectively mitigate overfitting, making it a robust solution for computational pathology applications. Keywords: Mamba · Computational Pathology · Whole Slide Images · Multiple Instance Learning.The paper "MambaMIL: Enhancing Long Sequence Modeling with Sequence Reordering in Computational Pathology" introduces a novel approach, MambaMIL, which integrates the Selective Scan Space State Sequential Model (Mamba) into Multiple Instance Learning (MIL) for computational pathology. The primary goal is to address the limitations of existing MIL methods in handling long sequences of instances and mitigating overfitting issues. MambaMIL leverages the capabilities of Mamba to model long sequences efficiently with linear complexity, while introducing the Sequence Reordering Mamba (SR-Mamba) to capture long-range dependencies and improve feature discrimination. SR-Mamba enhances the original Mamba by allowing interactions between instances in both sequential and transpositional orderings, thereby capturing more discriminative features. The method is evaluated on two challenging tasks—survival prediction and cancer subtyping—using nine diverse datasets. Extensive experiments demonstrate that MambaMIL outperforms state-of-the-art MIL methods, showing superior performance in terms of accuracy and stability. The proposed framework is also shown to effectively mitigate overfitting, making it a robust solution for computational pathology applications. Keywords: Mamba · Computational Pathology · Whole Slide Images · Multiple Instance Learning.