when dealing with patient-level labels like survival labels, one may encounter cases where multiple slides are available for a given patient
in that case, instead of concatenating regions from multiple slides into a single -- longer -- input sequence, pool from region to slide-level representation for each slide independently, then feed the sequence of slide-level representation into another (a 4th) Transformer block to pool these into a single patient-level representation
when dealing with patient-level labels like survival labels, one may encounter cases where multiple slides are available for a given patient
in that case, instead of concatenating regions from multiple slides into a single -- longer -- input sequence, pool from region to slide-level representation for each slide independently, then feed the sequence of slide-level representation into another (a 4th) Transformer block to pool these into a single patient-level representation