Error in training the optimal model

[shutiao888]

I want to know why, when training the optimal model, the loss decreases very slowly on the training log, the accuracy remains unchanged, and ultimately the malignant breast cancer is not recognized at all.Please help me clarify this.The following are some training logs.917/917 [==============================] - 1881s 2s/step - loss: 0.6928 - binary_accuracy: 0.5510 - val_loss: 0.6925 - val_binary_accuracy: 0.5507
Epoch 2/100
917/917 [==============================] - 99s 108ms/step - loss: 0.6922 - binary_accuracy: 0.5510 - val_loss: 0.6920 - val_binary_accuracy: 0.5507
Epoch 3/100
917/917 [==============================] - 99s 108ms/step - loss: 0.6917 - binary_accuracy: 0.5510 - val_loss: 0.6915 - val_binary_accuracy: 0.5507
Epoch 4/100
917/917 [==============================] - 101s 110ms/step - loss: 0.6913 - binary_accuracy: 0.5510 - val_loss: 0.6911 - val_binary_accuracy: 0.5507
Train for 917 steps, validate for 306 steps
Epoch 1/50
917/917 [==============================] - 119s 130ms/step - loss: 0.6879 - binary_accuracy: 0.5510 - val_loss: 0.6881 - val_binary_accuracy: 0.5507
Epoch 2/50
917/917 [==============================] - 107s 117ms/step - loss: 0.6879 - binary_accuracy: 0.5510 - val_loss: 0.6881 - val_binary_accuracy: 0.5507
Epoch 3/50
917/917 [==============================] - 107s 117ms/step - loss: 0.6879 - binary_accuracy: 0.5510 - val_loss: 0.6881 - val_binary_accuracy: 0.5507
Epoch 4/50
917/917 [==============================] - 107s 116ms/step - loss: 0.6879 - binary_accuracy: 0.5510 - val_loss: 0.6881 - val_binary_accuracy: 0.5507
precision recall f1-score support

  BENIGN       0.55      1.00      0.71       337                                                                   

MALIGNANT 0.00 0.00 0.00 275

accuracy                           0.55       612                                                                   

macro avg 0.28 0.50 0.36 612
weighted avg 0.30 0.55 0.39 612