I’m training a Attention Unet model on MRIs, and getting negative loss.
dropout 0.5
lr 0.001
batch size 16
loss function Dice loss
Optimizer Adam

what can be the reason of this.
and my model performance is not quite good what can I do?

Epoch 22/30
96/96 ━━━━━━━━━━━━━━━━━━━━ 47s 493ms/step – accuracy: 0.9529 – dice_coef: 0.9770 – iou_coef: 0.9603 – loss: 0.0231 – val_accuracy: 0.9632 – val_dice_coef: 0.5934 – val_iou_coef: 0.4448 – val_loss: 0.4072
Epoch 23/30
96/96 ━━━━━━━━━━━━━━━━━━━━ 47s 493ms/step – accuracy: 0.9555 – dice_coef: 0.9916 – iou_coef: 0.9891 – loss: 0.0084 – val_accuracy: 0.9633 – val_dice_coef: 0.5848 – val_iou_coef: 0.4418 – val_loss: 0.4147
Epoch 24/30
96/96 ━━━━━━━━━━━━━━━━━━━━ 47s 494ms/step – accuracy: 0.9550 – dice_coef: 1.0029 – iou_coef: 1.0106 – loss: -0.0028 – val_accuracy: 0.9637 – val_dice_coef: 0.5672 – val_iou_coef: 0.4215 – val_loss: 0.4320
Epoch 25/30
96/96 ━━━━━━━━━━━━━━━━━━━━ 47s 494ms/step – accuracy: 0.9554 – dice_coef: 1.0059 – iou_coef: 1.0198 – loss: -0.0058 – val_accuracy: 0.9634 – val_dice_coef: 0.4849 – val_iou_coef: 0.3360 – val_loss: 0.5142
Epoch 26/30
96/96 ━━━━━━━━━━━━━━━━━━━━ 47s 494ms/step – accuracy: 0.9554 – dice_coef: 1.0012 – iou_coef: 1.0112 – loss: -0.0012 – val_accuracy: 0.9625 – val_dice_coef: 0.7559 – val_iou_coef: 0.6388 – val_loss: 0.2439
Epoch 27/30
96/96 ━━━━━━━━━━━━━━━━━━━━ 47s 493ms/step – accuracy: 0.9538 – dice_coef: 1.0012 – iou_coef: 1.0082 – loss: -0.0012 – val_accuracy: 0.9636 – val_dice_coef: 0.3198 – val_iou_coef: 0.1962 – val_loss: 0.6804
Epoch 28/30
96/96 ━━━━━━━━━━━━━━━━━━━━ 47s 493ms/step – accuracy: 0.9560 – dice_coef: 1.0069 – iou_coef: 1.0195 – loss: -0.0069 – val_accuracy: 0.9620 – val_dice_coef: 0.6112 – val_iou_coef: 0.4666 – val_loss: 0.3895
Epoch 29/30
96/96 ━━━━━━━━━━━━━━━━━━━━ 47s 493ms/step – accuracy: 0.9562 – dice_coef: 1.0102 – iou_coef: 1.0272 – loss: -0.0102 – val_accuracy: 0.9622 – val_dice_coef: 0.7688 – val_iou_coef: 0.6625 – val_loss: 0.2318
Epoch 30/30
96/96 ━━━━━━━━━━━━━━━━━━━━ 47s 493ms/step – accuracy: 0.9553 – dice_coef: 1.0202 – iou_coef: 1.0499 – loss: -0.0204 – val_accuracy: 0.9637 – val_dice_coef: 0.5216 – val_iou_coef: 0.3763 – val_loss: 0.4775