--- title: 'CMFDNet: Dual Roles in Medical Imaging' url: https://www.emergentmind.com/topics/cmfdnet type: topic --- # CMFDNet: Dual Roles in Medical Imaging CMFDNet is an overloaded name in recent arXiv literature. In one usage, it denotes **“CMFDNet: Cross-Mamba and Feature Discovery Network for Polyp Segmentation”**, a Mamba/SSM-based medical image segmentation architecture for colonoscopy images that couples a Cross-Mamba Decoder (CMD), a Multi-Scale Aware (MSA) bridge, and a Feature Discovery (FD) head [2508.17729]. In another usage, the name is employed as a synonym for the complete network instantiating **Completed Feature Disentanglement Learning (CFDL)** for multimodal MRI classification, where the core components are Complete Feature Disentanglement (CFD) and Dynamic Mixture-of-Experts Fusion (DMF) [2407.04916]. The shared acronym masks two unrelated methodological lineages: one is a 2D dense prediction model built around Vision Mamba and decoder-side feature fusion, while the other is a multimodal classification framework based on disentangled latent representations and expert gating. ## 1. Nomenclature and scope The name **CMFDNet** is unambiguous only within the context of a given paper. In the polyp-segmentation paper, CMFDNet is the official model name and expands to **Cross-Mamba and Feature Discovery Network** [2508.17729]. In the multimodal MRI paper, the official method name is **Completed Feature Disentanglement Learning**, abbreviated **CFDL**; however, the provided overview explicitly states that **CMFDNet refers to the complete network that implements CFD + DMF** and is synonymous with the authors’ CFDL instantiation [2407.04916]. This naming overlap matters because the two systems differ in task, data modality, architectural primitives, and mathematical objectives. The segmentation CMFDNet operates on colonoscopy images resized to \(224\times224\), uses a pretrained **VMamba-Tiny** encoder, and produces a pixelwise segmentation map through a top-down decoder [2508.17729]. The MRI-classification CMFDNet/CFDL uses non-shared **3D ResNet-18** backbones per modality, decouples latent vectors into shared, specific, and partial-shared components, and feeds a fused representation to an MLP classifier [2407.04916]. A common misconception is therefore to treat CMFDNet as a single canonical architecture. The literature provided does not support that view. Instead, it documents two separate architectures that happen to share the same shorthand. ## 2. CMFDNet as a polyp-segmentation architecture In **“CMFDNet: Cross-Mamba and Feature Discovery Network for Polyp Segmentation”** [2508.17729], the target problem is automated colonic polyp segmentation in colonoscopy. The paper frames the task around three failure modes: significant variation in polyp shapes and sizes, indistinct boundaries between polyps and adjacent tissues, and under-detection of small polyps because information is lost during downsampling and decoding. The architecture is explicitly designed against these failure modes. The network uses **VMamba-Tiny (Vision Mamba)** as a pretrained encoder. Given an input image \(I\) of size \(H\times W\) with \(H=W=224\), the encoder produces a three-stage feature pyramid \(S_1,S_2,S_3\) at strides \(\{4,8,16\}\), corresponding to spatial sizes \(\{H/4,W/4\}\), \(\{H/8,W/8\}\), and \(\{H/16,W/16\}\) [2508.17729]. Each encoder feature is passed through an **MSA** module, and the resulting features feed a three-stage top-down **CMD** decoder. A downstream **FD** module then fuses decoder outputs across stages by multiplicative gating and residual addition, after which a final \(1\times1\) convolution produces the segmentation map. The architectural decomposition is as follows. | Component | Function | Key mechanism | |---|---|---| | MSA | Encoder-to-decoder bridge | Inverted bottleneck, multi-branch depthwise convolutions, ChannelShuffle | | CMD | Decoder | Row/column exchange, diagonal SS2D cross-scanning, GAB, separable convolutions | | FD | Prediction head fusion | Element-wise multiplication gating and residual fusion | The **MSA module** addresses scale diversity. It first applies **GAB** and pointwise convolution for channel expansion, then processes the feature map with parallel depthwise convolutions of sizes \(3\times3\), \(5\times5\), and \(7\times7\), and finally aggregates them through **Channel\_Shuffle** and a pointwise convolution: \[ O_i = PWC(GAB(S_i)) \] \[ O_i^1 = DWC_{3\times3}(O_i),\quad O_i^2 = DWC_{5\times5}(O_i),\quad O_i^3 = DWC_{7\times7}(O_i) \] \[ MSA_i = PWC(Channel\_Shuffle(O_i^1 + O_i^2 + O_i^3)) \] The stated rationale is that polyps vary widely in apparent scale under different camera-to-tissue distances and endoscope optics [2508.17729]. The **CMD module** is the decoder innovation. Its purpose is to reduce blurry boundaries by aligning deep semantic features with shallow local details through cross-scanning in complementary directions. At decoder stage \(i\), the current bridge feature \(MSA_i\) and the upsampled deeper decoder feature \(CMD_{i+1}^u\) undergo row and column exchange: \[ S_i^r, D_i^r = Row\_Exchange(MSA_i, CMD_{i+1}^{u}) \] \[ S_i^c, D_i^c = Column\_Exchange(MSA_i, CMD_{i+1}^{u}) \] Each exchanged tensor is then processed by **VMamba-style SS2D selective scanning** in four diagonal orders \((\searrow,\nearrow,\swarrow,\nwarrow)\): \[ R_i^1 = VSS\ Scan(S_i^r),\quad R_i^2 = VSS\ Scan(D_i^r) \] \[ C_i^1 = VSS\ Scan(S_i^c),\quad C_i^2 = VSS\ Scan(D_i^c) \] The fused output is refined by **GAB**, pointwise convolutions, and depthwise convolution: \[ B_i^1 = GAB(PWC(R_i^1 + R_i^2)),\quad B_i^2 = GAB(PWC(C_i^1 + C_i^2)) \] \[ CMD_i = PWC(DWC_{3\times3}(PWC(B_i^1 + B_i^2))) \] The paper’s explanation is that row/column exchange emphasizes axial contrasts before diagonal SS2D, thereby increasing sensitivity to edge orientation and reducing boundary blur [2508.17729]. The **FD module** targets small-polyp recovery. It introduces dependencies between decoder stages using transposed-convolution upsampling, element-wise multiplication, and residual addition: \[ G_{23} = DEC_{4\times4}(CMD_3)\odot CMD_2 + CMD_2 \] \[ output = DEC_{4\times4}(G_{23})\odot CMD_1 + CMD_1 \] Here, \(DEC_{4\times4}\) denotes a transposed convolution with kernel \(4\times4\) and padding \(1\) that doubles spatial resolution. The multiplicative term acts as a per-pixel affinity or gate from deeper semantics onto finer features [2508.17729]. ## 3. Mathematical basis and training protocol of the segmentation model The segmentation CMFDNet is grounded in the **selective state-space model (SSM)** formulation associated with Mamba. The paper states the general continuous-time linear SSM as \[ x_{t+1} = A x_t + B u_t,\quad y_t = C x_t + D u_t \] and describes **selective SSM** as introducing data-dependent gating: \[ s_t = \sigma(W_g u_t)\in(0,1),\quad \tilde{u}_t = s_t \odot u_t \] \[ x_{t+1} = A x_t + B \tilde{u}_t,\quad y_t = C x_t + D \tilde{u}_t \] In **SS2D**, 2D tensors are flattened into directional scan orders, and cross-scanning aggregates outputs from multiple directions. The broader stylized formulation provided includes eight directions, \[ y^{(d)} = \mathrm{SSM}_d(u),\quad d\in\{\rightarrow,\leftarrow,\downarrow,\uparrow,\searrow,\nearrow,\swarrow,\nwarrow\}, \] although this specific model uses four diagonal scans [2508.17729]. The **Global Attention Block (GAB)** is defined as a parallel channel/spatial attention mechanism with learnable trade-off \(\lambda\in(0,1)\): \[ W_{cs} = (1-\lambda)W_c' + \lambda W_s' \] \[ GAB_{out} = W_{cs}\odot M + M \] The paper contrasts this with serial CBAM and states that weighted parallel fusion preserves both global channel importance and spatial saliency [2508.17729]. Training uses deep supervision, but the paper does **not** specify the exact loss. The provided details recommend a standard compound loss matching the reported metrics, consisting of **Dice loss** and **binary cross-entropy**, summed over auxiliary side outputs and the final output. This is presented as a reproducibility recommendation rather than a claim that the paper explicitly used that exact formula [2508.17729]. The evaluation metrics explicitly reported are **Dice**, **IoU**, **E-measure (\(E_\xi\))**, **weighted F-measure (\(F_\beta^w\))**, **structure measure (\(S_\alpha\))**, and **MAE**. The datasets and training pipeline are clearly specified. Training uses **Kvasir-SEG (900 for training)** and **CVC-ClinicDB (550 for training)**; unused images from these datasets form visible test sets, while generalization is tested on **ETIS-LaribDB**, **CVC-ColonDB**, and **EndoScene**. Inputs are resized to \(224\times224\), augmentation includes random horizontal flip, rotation, and color perturbations, and optimization uses **AdamW** with initial learning rate \(1\times10^{-4}\), halved every 50 epochs for a total of 150 epochs, batch size 8, under **PyTorch** on an **NVIDIA A800 GPU** [2508.17729]. ## 4. Empirical results, ablations, and limitations of the segmentation model The segmentation CMFDNet is compared against **PraNet, FCBFormer, ECTransNet, Swin-UMamba, VM-UNet, and VM-UNetV2**. The paper states that it outperforms six SOTA methods, with the largest margins on the most difficult unseen datasets [2508.17729]. The headline quantitative results are dataset-specific. On **ETIS**, the model reports **mDice 81.85** and **mIoU 74.27**, with mDice exceeding the best SOTA baseline by **1.83%**; it also records top **\(F_\beta^w\) 77.53**, **\(S_\alpha\) 94.12**, **\(E_\xi\) 91.94**, and **MAE 1.35**. On **ColonDB**, it reports **mDice 83.05** and **mIoU 75.56**, improving the best SOTA mDice by **1.55%**, with top **\(F_\beta^w\) 81.44**, **\(S_\alpha\) 88.01**, **\(E_\xi\) 92.16**, and **MAE 2.99**. On **EndoScene**, **Kvasir**, and **ClinicDB**, the reported mDice values are **90.62**, **91.74**, and **93.36**, respectively, each with smaller positive margins over the best baseline [2508.17729]. The ablation studies isolate the contributions of the main modules. On **ColonDB**, the full model yields **mDice 83.05**, **mIoU 75.56**, and **MAE 2.99**. Removing **CMD** reduces performance to **mDice 77.12** and **mIoU 70.31**, with MAE rising to **3.67**; removing **MSA** yields **mDice 81.94** and **mIoU 74.77**; removing **FD** yields **mDice 82.50** and **mIoU 74.67**; and replacing **GAB** with **CBAM** yields **mDice 82.46** and **mIoU 74.89**. A parallel pattern is reported on **ETIS**, where removing CMD causes the largest drop, from **81.85/74.27** to **76.23/69.23** in mDice/mIoU [2508.17729]. The explicit conclusion given is that **CMD delivers the largest gains**, while **MSA contributes multi-scale robustness**, **FD enhances small-polyp detection**, and **GAB improves attention fusion over CBAM** in this design. The paper also offers a state-of-the-art comparison by architectural family. It characterizes **U-Net variants and CNN decoders** such as PraNet as strong in boundary refinement but limited by local receptive fields and weaker global context modeling; **hybrid CNN–Transformer models** such as FCBFormer and ECTransNet as improving multi-scale semantics but incurring quadratic attention costs or relying on edge heuristics; and **Mamba-based baselines** such as VM-UNet, VM-UNetV2, and Swin-UMamba as lacking the particular combination of row/column exchange, diagonal multi-scan fusion, multi-branch MSA bridging, and FD gating used here [2508.17729]. Several limitations are explicitly noted. The model does not contain dedicated **noise suppression**, even though colonoscopy commonly exhibits specular highlights and related artifacts. The paper further states that cross-dataset gains are strong, but larger domain shifts may require **domain adaptation**, **style augmentation**, or **test-time adaptation**. It also notes that **ultra-small targets remain challenging**, despite the FD module, and that directional SS2D contributions could be visualized to improve interpretability [2508.17729]. The reported complexity discussion is conceptual rather than empirical: parameter count, FLOPs, throughput, and FPS are **not reported**. ## 5. CMFDNet as Completed Feature Disentanglement Learning for multimodal MRI In **“Completed Feature Disentanglement Learning for Multimodal MRIs Analysis”** [2407.04916], the overview states that **CMFDNet refers to the complete network that implements CFD + DMF** and is synonymous with the authors’ **CFDL** instantiation. This usage belongs to multimodal MRI classification rather than image segmentation. The motivating problem is multimodal learning with more than two MRI sequences, such as **T1, T1C/T1Gd, T2, FLAIR, DWI/ADC,** and **PD**. The paper argues that conventional feature-disentanglement methods split modality representations into **modality-shared** and **modality-specific** components, then fuse them by concatenation or attention. When the number of modalities satisfies \(M\ge 3\), this two-way decomposition can lose information shared only by a proper subset of modalities. The proposed remedy is **Complete Feature Disentanglement (CFD)**, which augments shared and specific features with **modality-partial-shared features** [2407.04916]. Let \(M\) denote the number of modalities and let each modality-specific encoder \(B_m\) produce a latent vector \(f_m^i\in\mathbb{R}^{d_0}\) for sample \(i\). The paper defines: - **Modality-shared features** \[ F_m^i = E_{sh}(f_m^i),\qquad F^i = \frac{1}{M}\sum_{m=1}^M F_m^i \] - **Modality-specific features** \[ P_m^i = E_{sp}^{(m)}(f_m^i) \] - **Modality-partial-shared features** for subset \(S\subseteq\{1,\dots,M\}\) with \(|S|\ge 2\) \[ G_{S,m}^i = E_{ps}^{(S)}(f_m^i),\qquad G_S^i = \frac{1}{|S|}\sum_{m\in S} G_{S,m}^i \] The final set of decoupled features is \[ \mathcal{S}^i = \{F^i\}\cup\{P_m^i\}_{m=1}^M\cup\{G_S^i\}_{S\in\mathcal{P}}. \] Architecturally, the model comprises non-shared **3D ResNet-18** backbones per modality, single-FC disentanglement heads \(E_{sh}\), \(E_{sp}^{(m)}\), and \(E_{ps}^{(S)}\), followed by a **Dynamic Mixture-of-Experts Fusion (DMF)** block. In three-modality settings, \(\mathcal{P}\) contains all pairs; in four-modality settings, the model includes both pairs and triplets [2407.04916]. The disentanglement objective has three terms. First, **all-modality shared similarity**: \[ \mathcal{L}_{sh} = \sum_{j=1}^M\sum_{k=j+1}^M MSE(F_j,F_k) \] Second, **within-subset partial-shared similarity**: \[ \mathcal{L}_{ps} = \sum_{S\in\mathcal{P}}\sum_{j,k\in S,\;j