Reconstructing defect contours by inspecting internal magnetic flux leakage (MFL) in oil and gas pipelines provides significant data support for assessing pipeline integrity and represents a crucial step in MFL data analysis. Traditional methods often suffer from blurred edges and low structural similarity in multiple reconstructed images, making them inadequate for high-precision inspection. This paper proposes a three-dimensional (3-D) pipeline defect contour inversion method based on direction-aware multi-axis fusion and depth-gradient joint decoding (MFL-MAARN), which enhanced feature representation by fusing MFL signals with coordinate information. This work also designed a deep network architecture that combined a residual module with an attention gate mechanism and constructed a composite loss function combining MAE and the structural similarity index (SSIM). This achieved dual optimization aimed at both 3-D defect contour detail recovery and global structural reconstruction. The experimental results indicated that the MFL-MAARN model effectively reproduced the true defect morphology for rectangular, cylindrical, conical, regular, and irregular defects, accurately capturing defect depth information. The model also demonstrated high fitting accuracy for defect reconstruction from actual pipeline MFL internal inspection signals, further validating its generalization ability and robustness. This model provides efficient and accurate technical support for pipeline safety inspection.