A study introduces a lightweight convolutional neural network enhanced with transfer learning for multi-cancer detection using biomedical images. The architecture aims to reduce computational complexity while maintaining high classification performance for deployment in resource-constrained environments. Researchers evaluated the model on three tumor datasets comprising brain MRI and lung and kidney CT scans. The system achieved test accuracies of 90.85%, 98.64%, and 99.92% for brain, lung, and kidney cancer respectively via five-fold stratified cross-validation. Transfer learning was employed by pretraining on one cancer type and fine-tuning on others, requiring only 20 additional epochs to match scratch-trained models. The fine-tuning process updates the classification part of the CNN and takes approximately 0.014 seconds per image per epoch on an NVIDIA GeForce GTX 960. Comparative evaluations demonstrate that this model outperforms state-of-the-art architectures such as Xception, VGG16, VGG19, MobileNetV2, and DenseNet121.