3D CNN training accuracy does not increase - tensorflow

Question

I am training a medical dataset and the images are 3D arrays and I added a channel layer to run 3D CNN neural net. The images are only 25 and when I run the neural network the accuracy and loss do not change with epochs at all. after every epoch the accuracy is constant. I believe after every epochs the weights should change and accuracy should atleast increase by one percent but nothing happens here. I don't have a clue what is going wrong here.
#X shape=[25,100,100,100]
#input_shape_=[100,100,100]
tf.keras.backend.set_image_data_format('channels_last')
input_shape_=[resize_num,resize_num,resize_num,1]
model = tf.keras.models.Sequential([
tf.keras.layers.Conv3D(12, (2,2,2), activation='relu',input_shape=input_shape_),
tf.keras.layers.MaxPooling3D((2,2,2)),
tf.keras.layers.Conv3D(18, (2,2,2), activation='relu'),
tf.keras.layers.MaxPooling3D((3,3,3)),
tf.keras.layers.Conv3D(36, (3,3,3), activation='relu'),
tf.keras.layers.MaxPooling3D((3,3,3)),
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(200,'relu'),
tf.keras.layers.Dense(120,'relu'),
tf.keras.layers.Dense(60,'relu'),
tf.keras.layers.Dense(10,'relu'),
tf.keras.layers.Dense(1,'relu')
])
#model.summary()
model.compile(optimizer='Adam',loss='BinaryCrossentropy',metrics=['accuracy'])
model.fit(X_train,y,epochs=10,batch_size=1)
Epoch 1/10
25/25 [==============================] - 32s 1s/step - loss: 5.5460 - accuracy: 0.6400
Epoch 2/10
25/25 [==============================] - 32s 1s/step - loss: 5.5460 - accuracy: 0.6400
Epoch 3/10
25/25 [==============================] - 32s 1s/step - loss: 5.5460 - accuracy: 0.6400
Epoch 4/10
25/25 [==============================] - 32s 1s/step - loss: 5.5460 - accuracy: 0.6400
Epoch 5/10
25/25 [==============================] - 32s 1s/step - loss: 5.5460 - accuracy: 0.6400
Epoch 6/10
25/25 [==============================] - 32s 1s/step - loss: 5.5460 - accuracy: 0.6400
Epoch 7/10
25/25 [==============================] - 32s 1s/step - loss: 5.5460 - accuracy: 0.6400
Epoch 8/10
25/25 [==============================] - 32s 1s/step - loss: 5.5460 - accuracy: 0.6400
Epoch 9/10
25/25 [==============================] - 32s 1s/step - loss: 5.5460 - accuracy: 0.6400
Epoch 10/10
25/25 [==============================] - 32s 1s/step - loss: 5.5460 - accuracy: 0.6400
<keras.callbacks.History at 0x2ba76127cc8>

Answer 1

The last layer activation function is not proper for this problem as binary classification model, change it into sigmoid activation function
...
tf.keras.layers.Dense(1,'sigmoid')
...