AOT-biomaps 2.1.3__py3-none-any.whl → 2.9.233__py3-none-any.whl

AOT_biomaps/AOT_Recon/AOT_Optimizers/MAPEM.py

@@ -0,0 +1,456 @@
+from AOT_biomaps.AOT_Recon.ReconEnums import PotentialType
+from AOT_biomaps.AOT_Recon.AOT_PotentialFunctions.Quadratic import _Omega_QUADRATIC_CPU, _Omega_QUADRATIC_GPU
+from AOT_biomaps.AOT_Recon.AOT_PotentialFunctions.RelativeDifferences import _Omega_RELATIVE_DIFFERENCE_CPU, _Omega_RELATIVE_DIFFERENCE_GPU
+from AOT_biomaps.AOT_Recon.AOT_PotentialFunctions.Huber import _Omega_HUBER_PIECEWISE_CPU, _Omega_HUBER_PIECEWISE_GPU
+from AOT_biomaps.AOT_Recon.ReconTools import _build_adjacency_sparse, check_gpu_memory, calculate_memory_requirement
+from AOT_biomaps.Config import config
+
+import warnings
+import numpy as np
+import torch
+from tqdm import trange
+
+def MAPEM(
+    SMatrix, 
+    y, 
+    Omega,
+    beta,
+    delta=None,
+    gamma=None,
+    sigma=None,
+    numIterations=100,
+    isSavingEachIteration=True,
+    withTumor=True,
+    device=None,
+    max_saves=5000,
+    show_logs=True):
+    """
+    This method implements the MAPEM algorithm using either CPU or single-GPU PyTorch acceleration.
+    Multi-GPU and Multi-CPU modes are not implemented for this algorithm.
+    """
+    try:
+        tumor_str = "WITH" if withTumor else "WITHOUT"
+        # Auto-select device and method
+        if device is None:
+            if torch.cuda.is_available() and check_gpu_memory(config.select_best_gpu(), calculate_memory_requirement(SMatrix, y), show_logs=show_logs):
+                device = torch.device(f"cuda:{config.select_best_gpu()}")
+                use_gpu = True
+            else:
+                device = torch.device("cpu")
+                use_gpu = False
+        else:
+            use_gpu = device.type == "cuda"
+        # Dispatch to the appropriate implementation
+        if use_gpu:
+                return _MAPEM_GPU(SMatrix, y, Omega, beta, delta, gamma, sigma, numIterations, isSavingEachIteration, tumor_str, device, max_saves, show_logs=True)
+        else:
+                return _MAPEM_CPU(SMatrix, y, Omega, beta, delta, gamma, sigma, numIterations, isSavingEachIteration, tumor_str, max_saves, show_logs=True)
+    except Exception as e:
+        print(f"Error in MLEM: {type(e).__name__}: {e}")
+        return None, None
+
+def MAPEM_STOP(
+    SMatrix, 
+    y, 
+    Omega,
+    beta,
+    delta=None,
+    gamma=None,
+    sigma=None,
+    numIterations=100,
+    isSavingEachIteration=True,
+    withTumor=True,
+    device=None,
+    max_saves=5000,
+    show_logs=True):
+    """
+    This method implements the MAPEM_STOP algorithm using either CPU or single-GPU PyTorch acceleration.
+    Multi-GPU and Multi-CPU modes are not implemented for this algorithm.
+    """
+    try:
+        tumor_str = "WITH" if withTumor else "WITHOUT"
+        # Auto-select device and method
+        if device is None:
+            if torch.cuda.is_available() and check_gpu_memory(config.select_best_gpu(), calculate_memory_requirement(SMatrix, y), show_logs=show_logs):
+                device = torch.device(f"cuda:{config.select_best_gpu()}")
+                use_gpu = True
+            else:
+                device = torch.device("cpu")
+                use_gpu = False
+        else:
+            use_gpu = device.type == "cuda"
+        # Dispatch to the appropriate implementation
+        if use_gpu:
+                return _MAPEM_GPU_STOP(SMatrix, y, Omega, beta, delta, gamma, sigma, numIterations, isSavingEachIteration, tumor_str, device, max_saves, show_logs=True)
+        else:
+                return _MAPEM_CPU_STOP(SMatrix, y, Omega, beta, delta, gamma, sigma, numIterations, isSavingEachIteration, tumor_str, max_saves, show_logs=True)
+    except Exception as e:
+        print(f"Error in MLEM: {type(e).__name__}: {e}")
+        return None, None
+
+
+def _MAPEM_CPU_STOP(SMatrix, y, Omega, beta, delta, gamma, sigma, numIterations, isSavingEachIteration, tumor_str, max_saves, show_logs=True):
+    """
+    MAPEM version CPU simple - sans GPU - torch uniquement
+    """
+    try:
+        if not isinstance(Omega, PotentialType):
+            raise TypeError(f"Omega must be of type PotentialType, got {type(Omega)}")
+        if Omega == PotentialType.HUBER_PIECEWISE:
+            if delta is None:
+                raise ValueError("delta must be specified for HUBER_PIECEWISE potential type.")
+            if beta is None:
+                raise ValueError("beta must be specified for HUBER_PIECEWISE potential type.")
+        elif Omega == PotentialType.RELATIVE_DIFFERENCE:
+            if gamma is None:
+                raise ValueError("gamma must be specified for RELATIVE_DIFFERENCE potential type.")
+            if beta is None:
+                raise ValueError("beta must be specified for RELATIVE_DIFFERENCE potential type.")
+        elif Omega == PotentialType.QUADRATIC:
+            if sigma is None:
+                raise ValueError("sigma must be specified for QUADRATIC potential type.")
+            if beta is None:
+                raise ValueError("beta must be specified for QUADRATIC potential type.")
+        else:
+            raise ValueError(f"Unknown potential type: {Omega}")
+
+        SMatrix = torch.tensor(SMatrix, dtype=torch.float32)
+        y_tensor = torch.tensor(y, dtype=torch.float32)
+        T, Z, X, N = SMatrix.shape
+        A_flat = SMatrix.permute(0, 3, 1, 2).reshape(T * N, Z * X)
+        y_flat = y_tensor.reshape(-1)
+        I_0 = torch.ones((Z, X), dtype=torch.float32)
+        theta_list = [I_0]
+        results = [I_0.numpy()]
+        saved_indices = [0]
+        normalization_factor = SMatrix.sum(dim=(0, 3)).reshape(-1)
+        adj_index, adj_values = _build_adjacency_sparse(Z, X)
+
+        # Calculate save indices
+        if numIterations <= max_saves:
+            save_indices = list(range(numIterations))
+        else:
+            step = numIterations // max_saves
+            save_indices = list(range(0, numIterations, step))
+            if save_indices[-1] != numIterations - 1:
+                save_indices.append(numIterations - 1)
+
+        if Omega == PotentialType.HUBER_PIECEWISE:
+            description = f"AOT-BioMaps -- Bayesian Reconstruction Tomography: MAP-EM (Sparse HUBER β:{beta:.4f}, δ:{delta:.4f}) + STOP condition (penalized log-likelihood) ---- {tumor_str} TUMOR ---- processing on single CPU ----"
+        elif Omega == PotentialType.RELATIVE_DIFFERENCE:
+            description = f"AOT-BioMaps -- Bayesian Reconstruction Tomography: MAP-EM (Sparse RD β:{beta:.4f}, γ:{gamma:.4f}) + STOP condition (penalized log-likelihood) ---- {tumor_str} TUMOR ---- processing on single CPU ----"
+        elif Omega == PotentialType.QUADRATIC:
+            description = f"AOT-BioMaps -- Bayesian Reconstruction Tomography: MAP-EM (Sparse QUADRATIC β:{beta:.4f}, σ:{sigma:.4f}) + STOP condition (penalized log-likelihood) ---- {tumor_str} TUMOR ---- processing on single CPU ----"
+
+        iterator = trange(numIterations, desc=description) if show_logs else range(numIterations)
+        for it in iterator:
+            theta_p = theta_list[-1]
+            theta_p_flat = theta_p.reshape(-1)
+            q_flat = A_flat @ theta_p_flat
+            e_flat = (y_flat - q_flat) / (q_flat + torch.finfo(torch.float32).tiny)
+            c_flat = A_flat.T @ e_flat
+            if Omega == PotentialType.HUBER_PIECEWISE:
+                grad_U, hess_U, U_value = _Omega_HUBER_PIECEWISE_CPU(theta_p_flat, adj_index, adj_values, delta=delta)
+            elif Omega == PotentialType.RELATIVE_DIFFERENCE:
+                grad_U, hess_U, U_value = _Omega_RELATIVE_DIFFERENCE_CPU(theta_p_flat, adj_index, adj_values, gamma=gamma)
+            elif Omega == PotentialType.QUADRATIC:
+                grad_U, hess_U, U_value = _Omega_QUADRATIC_CPU(theta_p_flat, adj_index, adj_values, sigma=sigma)
+            denom = normalization_factor + theta_p_flat * beta * hess_U
+            num = theta_p_flat * (c_flat - beta * grad_U)
+            theta_next_flat = theta_p_flat + num / (denom + torch.finfo(torch.float32).tiny)
+            theta_next_flat = torch.clamp(theta_next_flat, min=0)
+            theta_next = theta_next_flat.reshape(Z, X)
+            theta_list[-1] = theta_next
+            if isSavingEachIteration and it in save_indices:
+                results.append(theta_next.numpy())
+                saved_indices.append(it+1)
+            log_likelihood = (y_flat * torch.log(q_flat + 1e-8) - (q_flat + 1e-8)).sum()
+            penalized_log_likelihood = log_likelihood - beta * U_value
+            if (it + 1) % 100 == 0:
+                print(f"Iter {it+1}: logL={log_likelihood:.3e}, U={U_value:.3e}, penalized logL={penalized_log_likelihood:.3e}")
+
+        if isSavingEachIteration:
+            return results, saved_indices
+        else:
+            return results[-1], None
+    except Exception as e:
+        print(f"An error occurred in _MAPEM_CPU_STOP: {e}")
+        return None, None
+
+def _MAPEM_GPU_STOP(SMatrix, y, Omega, beta, delta, gamma, sigma, numIterations, isSavingEachIteration, tumor_str, device, max_saves, show_logs=True):
+    """
+    Maximum A Posteriori (MAP) estimation for Bayesian reconstruction.
+    This method computes the MAP estimate of the parameters given the data.
+    """
+    try:
+        if not isinstance(Omega, PotentialType):
+            raise TypeError(f"Omega must be of type PotentialType, got {type(Omega)}")
+        if Omega == PotentialType.HUBER_PIECEWISE:
+            if delta is None:
+                raise ValueError("delta must be specified for HUBER_PIECEWISE potential type.")
+            if beta is None:
+                raise ValueError("beta must be specified for HUBER_PIECEWISE potential type.")
+        elif Omega == PotentialType.RELATIVE_DIFFERENCE:
+            if gamma is None:
+                raise ValueError("gamma must be specified for RELATIVE_DIFFERENCE potential type.")
+            if beta is None:
+                raise ValueError("beta must be specified for RELATIVE_DIFFERENCE potential type.")
+        elif Omega == PotentialType.QUADRATIC:
+            if sigma is None:
+                raise ValueError("sigma must be specified for QUADRATIC potential type.")
+            if beta is None:
+                raise ValueError("beta must be specified for QUADRATIC potential type.")
+        else:
+            raise ValueError(f"Unknown potential type: {Omega}")
+
+        A_matrix_torch = torch.tensor(SMatrix, dtype=torch.float32).to(device)
+        y_torch = torch.tensor(y, dtype=torch.float32).to(device)
+        T, Z, X, N = SMatrix.shape
+        J = Z * X
+        A_flat = A_matrix_torch.permute(0, 3, 1, 2).reshape(T * N, Z * X)
+        y_flat = y_torch.reshape(-1)
+        I_0 = torch.ones((Z, X), dtype=torch.float32, device=device)
+        matrix_theta_torch = [I_0]
+        matrix_theta_from_gpu_MAPEM = [I_0.cpu().numpy()]
+        saved_indices = [0]
+        normalization_factor = A_matrix_torch.sum(dim=(0, 3))
+        normalization_factor_flat = normalization_factor.reshape(-1)
+        previous = -np.inf
+        nb_false_successive = 0
+        adj_index, adj_values = _build_adjacency_sparse(Z, X, device=device)
+
+        # Calculate save indices
+        if numIterations <= max_saves:
+            save_indices = list(range(numIterations))
+        else:
+            step = numIterations // max_saves
+            save_indices = list(range(0, numIterations, step))
+            if save_indices[-1] != numIterations - 1:
+                save_indices.append(numIterations - 1)
+
+        if Omega == PotentialType.HUBER_PIECEWISE:
+            description = f"AOT-BioMaps -- Bayesian Reconstruction Tomography: MAP-EM (Sparse HUBER β:{beta:.4f}, δ:{delta:.4f}) + STOP condition (penalized log-likelihood) ---- {tumor_str} TUMOR ---- processing on single GPU no.{torch.cuda.current_device()} ----"
+        elif Omega == PotentialType.RELATIVE_DIFFERENCE:
+            description = f"AOT-BioMaps -- Bayesian Reconstruction Tomography: MAP-EM (Sparse RD β:{beta:.4f}, γ:{gamma:.4f}) + STOP condition (penalized log-likelihood) ---- {tumor_str} TUMOR ---- processing on single GPU no.{torch.cuda.current_device()} ----"
+        elif Omega == PotentialType.QUADRATIC:
+            description = f"AOT-BioMaps -- Bayesian Reconstruction Tomography: MAP-EM (Sparse QUADRATIC β:{beta:.4f}, σ:{sigma:.4f}) + STOP condition (penalized log-likelihood) ---- {tumor_str} TUMOR ---- processing on single GPU no.{torch.cuda.current_device()} ----"
+
+        iterator = trange(numIterations, desc=description) if show_logs else range(numIterations)
+        for it in iterator:
+            theta_p = matrix_theta_torch[-1]
+            theta_p_flat = theta_p.reshape(-1)
+            q_flat = A_flat @ theta_p_flat
+            e_flat = (y_flat - q_flat) / (q_flat + torch.finfo(torch.float32).tiny)
+            c_flat = A_flat.T @ e_flat
+            if Omega == PotentialType.HUBER_PIECEWISE:
+                grad_U, hess_U, U_value = _Omega_HUBER_PIECEWISE_GPU(theta_p_flat, adj_index, adj_values, device=device, delta=delta)
+            elif Omega == PotentialType.RELATIVE_DIFFERENCE:
+                grad_U, hess_U, U_value = _Omega_RELATIVE_DIFFERENCE_GPU(theta_p_flat, adj_index, adj_values, device=device, gamma=gamma)
+            elif Omega == PotentialType.QUADRATIC:
+                grad_U, hess_U, U_value = _Omega_QUADRATIC_GPU(theta_p_flat, adj_index, adj_values, device=device, sigma=sigma)
+            else:
+                raise ValueError(f"Unknown potential type: {Omega}")
+            denom = normalization_factor_flat + theta_p_flat * beta * hess_U
+            num = theta_p_flat * (c_flat - beta * grad_U)
+            theta_p_plus_1_flat = theta_p_flat + num / (denom + torch.finfo(torch.float32).tiny)
+            theta_p_plus_1_flat = torch.clamp(theta_p_plus_1_flat, min=0)
+            theta_next = theta_p_plus_1_flat.reshape(Z, X)
+            matrix_theta_torch[-1] = theta_next
+            if isSavingEachIteration and it in save_indices:
+                matrix_theta_from_gpu_MAPEM.append(theta_next.cpu().numpy())
+                saved_indices.append(it+1)
+            log_likelihood = (y_flat * (torch.log(q_flat + torch.finfo(torch.float32).tiny)) - (q_flat + torch.finfo(torch.float32).tiny)).sum()
+            penalized_log_likelihood = log_likelihood - beta * U_value
+            if it == 0 or (it + 1) % 100 == 0:
+                current = penalized_log_likelihood.item()
+                if current <= previous:
+                    nb_false_successive += 1
+                else:
+                    nb_false_successive = 0
+                print(f"Iter {it + 1}: lnL without term ln(m_i !) inside={log_likelihood.item():.8e}, Gibbs energy function U={U_value.item():.4e}, penalized lnL without term ln(m_i !) inside={penalized_log_likelihood.item():.8e}, p lnL (current {current:.8e} - previous {previous:.8e} > 0)={(current - previous > 0)}, nb_false_successive={nb_false_successive}")
+                previous = current
+
+        del A_matrix_torch, y_torch, A_flat, y_flat, I_0, normalization_factor, normalization_factor_flat
+        torch.cuda.empty_cache()
+        if isSavingEachIteration:
+            return matrix_theta_from_gpu_MAPEM, saved_indices
+        else:
+            return matrix_theta_from_gpu_MAPEM[-1], None
+    except Exception as e:
+        print(f"An error occurred in _MAPEM_GPU_STOP: {e}")
+        del A_matrix_torch, y_torch, A_flat, y_flat, I_0, normalization_factor, normalization_factor_flat
+        torch.cuda.empty_cache()
+        return None, None
+
+def _MAPEM_CPU(SMatrix, y, Omega, beta, delta, gamma, sigma, numIterations, isSavingEachIteration, tumor_str, max_saves, show_logs=True):
+    try:
+        if not isinstance(Omega, PotentialType):
+            raise TypeError(f"Omega must be of type PotentialType, got {type(Omega)}")
+        if Omega == PotentialType.HUBER_PIECEWISE:
+            if delta is None:
+                raise ValueError("delta must be specified for HUBER_PIECEWISE potential type.")
+            if beta is None:
+                raise ValueError("beta must be specified for HUBER_PIECEWISE potential type.")
+        elif Omega == PotentialType.RELATIVE_DIFFERENCE:
+            if gamma is None:
+                raise ValueError("gamma must be specified for RELATIVE_DIFFERENCE potential type.")
+            if beta is None:
+                raise ValueError("beta must be specified for RELATIVE_DIFFERENCE potential type.")
+        elif Omega == PotentialType.QUADRATIC:
+            if sigma is None:
+                raise ValueError("sigma must be specified for QUADRATIC potential type.")
+            if beta is None:
+                raise ValueError("beta must be specified for QUADRATIC potential type.")
+        else:
+            raise ValueError(f"Unknown potential type: {Omega}")
+
+        T, Z, X, N = SMatrix.shape
+        A_flat = np.transpose(SMatrix, (0, 3, 1, 2)).reshape(T * N, Z * X)
+        y_flat = y.reshape(-1)
+        theta_0 = np.ones((Z, X), dtype=np.float32)
+        matrix_theta_np = [theta_0]
+        I_reconMatrix = [theta_0.copy()]
+        saved_indices = [0]
+        normalization_factor = SMatrix.sum(axis=(0, 3))
+        normalization_factor_flat = normalization_factor.reshape(-1)
+        adj_index, adj_values = _build_adjacency_sparse(Z, X)
+
+        # Calculate save indices
+        if numIterations <= max_saves:
+            save_indices = list(range(numIterations))
+        else:
+            step = numIterations // max_saves
+            save_indices = list(range(0, numIterations, step))
+            if save_indices[-1] != numIterations - 1:
+                save_indices.append(numIterations - 1)
+
+        if Omega == PotentialType.HUBER_PIECEWISE:
+            description = f"AOT-BioMaps -- Bayesian Reconstruction Tomography: MAP-EM (Sparse HUBER β:{beta:.4f}, δ:{delta:.4f}) ---- {tumor_str} TUMOR ---- processing on single CPU ----"
+        elif Omega == PotentialType.RELATIVE_DIFFERENCE:
+            description = f"AOT-BioMaps -- Bayesian Reconstruction Tomography: MAP-EM (Sparse RD β:{beta:.4f}, γ:{gamma:.4f}) ---- {tumor_str} TUMOR ---- processing on single CPU ----"
+        elif Omega == PotentialType.QUADRATIC:
+            description = f"AOT-BioMaps -- Bayesian Reconstruction Tomography: MAP-EM (Sparse QUADRATIC β:{beta:.4f}, σ:{sigma:.4f}) ---- {tumor_str} TUMOR ---- processing on single CPU ----"
+
+        iterator = trange(numIterations, desc=description) if show_logs else range(numIterations)
+        for it in iterator:
+            theta_p = matrix_theta_np[-1]
+            theta_p_flat = theta_p.reshape(-1)
+            q_flat = A_flat @ theta_p_flat
+            e_flat = (y_flat - q_flat) / (q_flat + np.finfo(np.float32).tiny)
+            c_flat = A_flat.T @ e_flat
+            if Omega == PotentialType.HUBER_PIECEWISE:
+                grad_U, hess_U, _ = _Omega_HUBER_PIECEWISE_CPU(theta_p_flat, adj_index, adj_values, delta=delta)
+            elif Omega == PotentialType.RELATIVE_DIFFERENCE:
+                grad_U, hess_U, _ = _Omega_RELATIVE_DIFFERENCE_CPU(theta_p_flat, adj_index, adj_values, gamma=gamma)
+            elif Omega == PotentialType.QUADRATIC:
+                grad_U, hess_U, _ = _Omega_QUADRATIC_CPU(theta_p_flat, adj_index, adj_values, sigma=sigma)
+            denom = normalization_factor_flat + theta_p_flat * beta * hess_U
+            num = theta_p_flat * (c_flat - beta * grad_U)
+            theta_p_plus_1_flat = theta_p_flat + num / (denom + np.finfo(np.float32).tiny)
+            theta_p_plus_1_flat = np.clip(theta_p_plus_1_flat, 0, None)
+            theta_next = theta_p_plus_1_flat.reshape(Z, X)
+            matrix_theta_np.append(theta_next)
+            if isSavingEachIteration and it in save_indices:
+                I_reconMatrix.append(theta_next.copy())
+                saved_indices.append(it+1)
+
+        if isSavingEachIteration:
+            return I_reconMatrix, saved_indices
+        else:
+            return I_reconMatrix[-1], None
+    except Exception as e:
+        print(f"An error occurred in _MAPEM_CPU: {e}")
+        return None, None
+
+def _MAPEM_GPU(SMatrix, y, Omega, beta, delta, gamma, sigma, numIterations, isSavingEachIteration, tumor_str, device, max_saves, show_logs=True):
+    """
+    Maximum A Posteriori (MAP) estimation for Bayesian reconstruction using GPU.
+    This method computes the MAP estimate of the parameters given the data.
+    """
+    try:
+        if not isinstance(Omega, PotentialType):
+            raise TypeError(f"Omega must be of type PotentialType, got {type(Omega)}")
+        if Omega == PotentialType.HUBER_PIECEWISE:
+            if delta is None:
+                raise ValueError("delta must be specified for HUBER_PIECEWISE potential type.")
+            if beta is None:
+                raise ValueError("beta must be specified for HUBER_PIECEWISE potential type.")
+        elif Omega == PotentialType.RELATIVE_DIFFERENCE:
+            if gamma is None:
+                raise ValueError("gamma must be specified for RELATIVE_DIFFERENCE potential type.")
+            if beta is None:
+                raise ValueError("beta must be specified for RELATIVE_DIFFERENCE potential type.")
+        elif Omega == PotentialType.QUADRATIC:
+            if sigma is None:
+                raise ValueError("sigma must be specified for QUADRATIC potential type.")
+            if beta is None:
+                raise ValueError("beta must be specified for QUADRATIC potential type.")
+        else:
+            raise ValueError(f"Unknown potential type: {Omega}")
+
+        A_matrix_torch = torch.tensor(SMatrix, dtype=torch.float32).to(device)
+        y_torch = torch.tensor(y, dtype=torch.float32).to(device)
+        T, Z, X, N = SMatrix.shape
+        J = Z * X
+        A_flat = A_matrix_torch.permute(0, 3, 1, 2).reshape(T * N, Z * X)
+        y_flat = y_torch.reshape(-1)
+        theta_0 = torch.ones((Z, X), dtype=torch.float32, device=device)
+        matrix_theta_torch = [theta_0]
+        I_reconMatrix = [theta_0.cpu().numpy()]
+        saved_indices = [0]
+        normalization_factor = A_matrix_torch.sum(dim=(0, 3))
+        normalization_factor_flat = normalization_factor.reshape(-1)
+        adj_index, adj_values = _build_adjacency_sparse(Z, X, device=device)
+
+        # Calculate save indices
+        if numIterations <= max_saves:
+            save_indices = list(range(numIterations))
+        else:
+            step = numIterations // max_saves
+            save_indices = list(range(0, numIterations, step))
+            if save_indices[-1] != numIterations - 1:
+                save_indices.append(numIterations - 1)
+
+        if Omega == PotentialType.HUBER_PIECEWISE:
+            description = f"AOT-BioMaps -- Bayesian Reconstruction Tomography: MAP-EM (Sparse HUBER β:{beta:.4f}, δ:{delta:.4f}) ---- {tumor_str} TUMOR ---- processing on single GPU no.{torch.cuda.current_device()} ----"
+        elif Omega == PotentialType.RELATIVE_DIFFERENCE:
+            description = f"AOT-BioMaps -- Bayesian Reconstruction Tomography: MAP-EM (Sparse RD β:{beta:.4f}, γ:{gamma:.4f}) ---- {tumor_str} TUMOR ---- processing on single GPU no.{torch.cuda.current_device()} ----"
+        elif Omega == PotentialType.QUADRATIC:
+            description = f"AOT-BioMaps -- Bayesian Reconstruction Tomography: MAP-EM (Sparse QUADRATIC σ:{sigma:.4f}) ---- {tumor_str} TUMOR ---- processing on single GPU no.{torch.cuda.current_device()} ----"
+
+        iterator = trange(numIterations, desc=description) if show_logs else range(numIterations)
+        for it in iterator:
+            theta_p = matrix_theta_torch[-1]
+            theta_p_flat = theta_p.reshape(-1)
+            q_flat = A_flat @ theta_p_flat
+            e_flat = (y_flat - q_flat) / (q_flat + torch.finfo(torch.float32).tiny)
+            c_flat = A_flat.T @ e_flat
+            if Omega == PotentialType.HUBER_PIECEWISE:
+                grad_U, hess_U, _ = _Omega_HUBER_PIECEWISE_GPU(theta_p_flat, adj_index, adj_values, device=device, delta=delta)
+            elif Omega == PotentialType.RELATIVE_DIFFERENCE:
+                grad_U, hess_U, _ = _Omega_RELATIVE_DIFFERENCE_GPU(theta_p_flat, adj_index, adj_values, device=device, gamma=gamma)
+            elif Omega == PotentialType.QUADRATIC:
+                grad_U, hess_U, _ = _Omega_QUADRATIC_GPU(theta_p_flat, adj_index, adj_values, device=device, sigma=sigma)
+            else:
+                raise ValueError(f"Unknown potential type: {Omega}")
+            denom = normalization_factor_flat + theta_p_flat * beta * hess_U
+            num = theta_p_flat * (c_flat - beta * grad_U)
+            theta_p_plus_1_flat = theta_p_flat + num / (denom + torch.finfo(torch.float32).tiny)
+            theta_p_plus_1_flat = torch.clamp(theta_p_plus_1_flat, min=0)
+            theta_next = theta_p_plus_1_flat.reshape(Z, X)
+            matrix_theta_torch.append(theta_next)
+            if isSavingEachIteration and it in save_indices:
+                I_reconMatrix.append(theta_next.cpu().numpy())
+                saved_indices.append(it+1)
+
+        del A_matrix_torch, y_torch, A_flat, y_flat, theta_0, normalization_factor, normalization_factor_flat
+        torch.cuda.empty_cache()
+        if isSavingEachIteration:
+            return I_reconMatrix, saved_indices
+        else:
+            return I_reconMatrix[-1], None
+    except Exception as e:
+        print(f"An error occurred in _MAPEM_GPU: {e}")
+        del A_matrix_torch, y_torch, A_flat, y_flat, theta_0, normalization_factor, normalization_factor_flat
+        torch.cuda.empty_cache()
+        return None, None